Datasets:

nineninesix
/

multilingual-tts-benchmark

utt string	lang string	subset string	source string	cv_version string	prompt_text string	prompt_dur float64	prompt_sr_orig int64	prompt_dur_bin string	prompt_cv_path string	prompt_split_origin string	speaker_id string	speaker_gender string	speaker_gender_source string	speaker_age string	speaker_accent_label string	speaker_n_in_subset int64	qc_vad_speech_ratio float64	qc_lead_sil float64	qc_trail_sil float64	qc_clip_rate float64	qc_bandwidth_hz float64	qc_spk_centroid_dist float32	qc_dnsmos_ovrl float32	qc_dnsmos_sig float32	qc_dnsmos_bak float32	text string	text_norm string	n_words int64	n_chars int64	len_bin string	phones string	n_phones int64	punct_type string	text_cv_path string	text_split_origin string	has_gt bool	gt_dur float64	gt_speaker_id string	gt_same_speaker bool	gt_gender string	sim_ref_audio audio	has_sim_ref bool	sim_ref_dur float64	sim_ref_text string	category string	tags list	has_digit bool	has_abbrev bool	has_foreign bool	sim_ref_dur_bin string	anchor_asr string	anchor_wer float32	anchor_cer float32	anchor_hyp string	anchor_subs int32	anchor_dels int32	anchor_ins int32	anchor_n_ref_words int32	anchor_cer_err int32	anchor_n_ref_chars int32	anchor_wer_mms float32	anchor_cer_mms float32	anchor_subs_mms int32	anchor_dels_mms int32	anchor_ins_mms int32	anchor_cer_err_mms int32	anchor_wer_scribe float32	anchor_cer_scribe float32	anchor_subs_scribe int32	anchor_dels_scribe int32	anchor_ins_scribe int32	anchor_cer_err_scribe int32	anchor_hyp_scribe string	anchor_sim_wavlm_sv float32	anchor_sim_wavlm_ft float32	anchor_sim_ecapa float32	gt_qc_dnsmos_ovrl float32	gt_qc_dnsmos_sig float32	gt_qc_dnsmos_bak float32	gt_qc_hf_energy_db float32	gt_qc_vad_speech_ratio float32	gt_qc_lead_sil float32	gt_qc_trail_sil float32	gt_qc_fails_prompt_qc bool	simref_qc_dnsmos_ovrl float32	simref_qc_dnsmos_sig float32	simref_qc_dnsmos_bak float32	simref_qc_clip_rate float32	simref_qc_hf_energy_db float32	simref_qc_vad_speech_ratio float32	simref_qc_lead_sil float32	simref_qc_trail_sil float32	simref_qc_fails_prompt_qc bool	qc_hf_energy_db float64
en-US_0001	en-US	main	cv17	17.0	There are at least two buses an hour towards Lancaster and Garstang.	3.808	48,000	3.7-4.5	common_voice_en_20582508.mp3	train	9f233d229308	male	cv	twenties	United States English	1	0.973739	0.61	0.914	0	-16.79018	0.973833	3.053883	3.408383	3.866694	Many scientists work their entire careers on this, and you say you have solved this in a single afternoon?	many scientists work their entire careers on this and you say you have solved this in a single afternoon	19	106	17+	m ɛ n i s a ɪ ə n t ɪ s t s w ɜ k ð ɛ ɹ ɛ n t a ɪ ɚ k ɚ ɹ ɪ ɹ z ɔ n ð ɪ s æ n d j u s e ɪ j u h æ v s ɑ l v d ð ɪ s ɪ n ɐ s ɪ ŋ ɡ ə l æ f t ɚ n u n	74	question	common_voice_en_18313200.mp3	test	true	6.432	fabbb092024f	false	male		true	2.336	Also, occasionally by Monty Munro.	general	[]	false	false	false	<3.5	whisper	0	0	many scientists work their entire careers on this and you say you have solved this in a single afternoon	0	0	0	19	0	104	0	0	0	0	0	0	0	0	0	0	0	0	many scientists work their entire careers on this and you say you have solved this in a single afternoon	0.913709	0.54596	0.548011	3.416343	3.675524	4.101205	-19.496309	0.909515	0.066	0	false	2.811669	3.365075	3.628996	0	-23.027567	0.971747	0.066	0	false	-16.79018
en-US_0002	en-US	main	cv17	17.0	Rather than take a step backward I shall thus succeed in taking one forward.	4.544	48,000	>=4.5	common_voice_en_18812547.mp3	train	895f2d4bf294	female	cv	thirties	United States English	2	0.977993	0.802	0.922	0.00001	-17.932814	0.97887	3.01907	3.557075	3.55249	That's why you collapsed?	that's why you collapsed	4	25	3-5	ð æ t s w a ɪ j u k ə l æ p s t	16	question	common_voice_en_17357574.mp3	test	true	1.376	f2b5dd789466	false	male		true	3.296	The town is a receival site for Cooperative Bulk Handling.	general	[]	false	false	false	<3.5	whisper	0	0	that's why you collapsed	0	0	0	4	0	24	0.5	0.083333	2	0	0	2	0	0	0	0	0	0	that's why you collapsed	0.929489	0.574692	0.534982	1.598097	2.626859	1.820628	-8.684204	0.975291	0.034	0	false	3.064922	3.410398	3.876479	0	-18.155439	0.979976	0.066	0	false	-17.932814
en-US_0003	en-US	main	cv17	17.0	Asymptomatic infection may be much more common among those infected in Europe.	4.832	32,000	>=4.5	common_voice_en_28132703.mp3	train	92e8b507ff4a	female	cv	twenties	United States English	2	0.979305	0.93	0.998	0	-26.951704	0.993042	2.82525	3.326467	3.556599	Wickham has used the technique for The Waterboys song Is She Conscious?	wickham has used the technique for the waterboys song is she conscious	12	71	10-12	w ɪ k æ m h ɐ z j u z d ð ə t ɛ k n i k f ɚ ð ə w ɔ ɾ ɚ b ɔ ɪ z s ɔ ŋ ɪ z ʃ i k ɑ n ʃ ə s	45	question	common_voice_en_27697437.mp3	test	true	5.728	baa6d53d4b0b	false	male		true	3.328	The use of quarter-tones requires a different embouchure.	general	[]	false	false	false	<3.5	whisper	0.416667	0.128571	we can have use the technique for the waterboy song is she conscious	4	0	1	12	9	70	0.5	0.142857	4	0	2	10	0	0	0	0	0	0	wickham has used the technique for the waterboys song is she conscious	0.958893	0.764994	0.6925	2.347444	3.191955	2.737565	-24.157827	0.976606	0.034	0	false	3.156523	3.44157	4.044859	0	-26.316648	0.980168	0.066	0	false	-26.951704
en-US_0004	en-US	main	cv17	17.0	He speaks with a lilting, musical intonation.	2.432	32,000	<3.7	common_voice_en_39624168.mp3	test	f96eaeb9028e	unknown	unknown	unknown	United States English	1	0.958882	1.442	0.906	0	-39.183516	1	2.946978	3.192843	4.006768	He is also known for writing Flash!	he is also known for writing flash	7	35	6-9	h i ɪ z ɔ l s o ʊ n o ʊ n f ɔ ɹ ɹ a ɪ ɾ ɪ ŋ f l æ ʃ	26	exclamation	common_voice_en_19933225.mp3	test	true	2.528	adf91d0e7e4a	false	male	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0	0	he is also known for writing flash	0	0	0	7	0	34	0.142857	0.029412	1	0	0	1	0	0	0	0	0	0	he is also known for writing flash	null	null	null	2.063095	3.261345	2.024566	-9.322443	0.986551	0.034	0	false	null	null	null	null	null	null	null	null	null	-39.183516
en-US_0005	en-US	main	cv17	17.0	The ideas on which it was based had ceased to be understood.	2.912	32,000	<3.7	common_voice_en_38278567.mp3	test	475c0c83b9f7	male	cv	thirties	United States English,Appalachian English	1	0.965659	0.674	1.986	0	-4.82564	0.973117	2.06994	3.059229	2.342568	Why don't you say something?	why don't you say something	5	28	3-5	w a ɪ d o ʊ n t j u s e ɪ s ʌ m θ ɪ ŋ	19	question	common_voice_en_17298838.mp3	test	true	1.472	b442ad42e929	false	male		true	2.592	She may not scratch her head or her body with her hands.	general	[]	false	false	false	<3.5	whisper	0	0	why don't you say something	0	0	0	5	0	27	0	0	0	0	0	0	0	0	0	0	0	0	why don't you say something	0.892131	0.458108	0.4459	2.889942	3.346126	3.655355	-12.713055	0.955163	0.066	0	false	1.614832	2.436311	1.806247	0	-13.359949	0.974537	0.066	0	false	-4.82564
en-US_0006	en-US	main	cv17	17.0	A rare, mirror image of the Segond fracture has also been described.	4.192	32,000	3.7-4.5	common_voice_en_37420122.mp3	test	6a8748bf754b	unknown	unknown	unknown	United States English	1	0.976145	0.578	1.198	0	-14.60871	1	2.502943	3.403867	2.83764	Shall I meet you in the concourse then?	shall i meet you in the concourse then	8	39	6-9	ʃ æ l a ɪ m i t j u ɪ n ð ə k ɑ ŋ k o ɹ s ð ɛ n	24	question	common_voice_en_17300476.mp3	test	true	2.432	b718c9253d62	false	male	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0	0	shall i meet you in the concourse then	0	0	0	8	0	38	0	0	0	0	0	0	0	0	0	0	0	0	shall i meet you in the concourse then	null	null	null	2.07042	3.268357	1.976736	-18.962753	0.999178	0.002	0	false	null	null	null	null	null	null	null	null	null	-14.60871
en-US_0007	en-US	main	cv17	17.0	I can reassure you, that we won't retreat until the oil until the peace is secured.	5.632	48,000	>=4.5	common_voice_en_17357595.mp3	test	0f71516b0970	male	cv	fourties	United States English	1	0.958807	0.738	1.074	0	-29.974683	1	3.360287	3.588934	4.11282	Is that how your niece began?	is that how your niece began	6	29	6-9	ɪ z ð æ t h a ʊ j ʊ ɹ n i s b ɪ ɡ æ n	19	question	common_voice_en_18366209.mp3	test	true	2.304	b7c15c14cf7f	false	female	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0	0	is that how your niece began	0	0	0	6	0	28	0	0	0	0	0	0	0.166667	0.071429	1	0	0	2	is that how your niche began	null	null	null	2.034134	3.313418	1.874005	-9.128964	0.971354	0.066	0	false	null	null	null	null	null	null	null	null	null	-29.974683
en-US_0008	en-US	main	cv17	17.0	It is designed to behead and fillet fish.	3.36	48,000	<3.7	common_voice_en_19848988.mp3	train	3f5d92c3cbb1	female	cv	fourties	United States English	2	0.970238	0.418	0.978	0	-23.657236	0.982401	2.446767	3.30042	2.889421	Where are the keys?	where are the keys	4	19	3-5	w ɛ ɹ ɑ ɹ ð ə k i z	10	question	common_voice_en_442796.mp3	test	true	2.784	be4fcf5645b8	false	male		true	2.112	Sometimes, no life is better.	general	[]	false	false	false	<3.5	whisper	0	0	where are the keys	0	0	0	4	0	18	0.75	0.277778	3	0	0	5	0	0	0	0	0	0	where are the keys	0.944554	0.606069	0.546206	2.523791	2.949472	3.549792	-15.089808	0.859914	0.066	0	false	2.359682	3.484429	2.392828	0	-15.954675	0.96875	0.066	0	false	-23.657236
en-US_0009	en-US	main	cv17	17.0	His daughter was Helen Vlachos.	2.016	32,000	<3.7	common_voice_en_31354431.mp3	test	d605c8a2c04c	unknown	unknown	unknown	United States English	1	0.950397	1.026	1.486	0	-23.73422	1	2.972338	3.568681	3.408793	Fite has said that she found the character unappealing: I thought, 'A cat?	fite has said that she found the character unappealing i thought 'a cat	13	74	13-16	f a ɪ t h ɐ z s ɛ d ð æ t ʃ i f a ʊ n d ð ə k æ ɹ ɪ k t ɚ ɹ ʌ n ɐ p i l ɪ ŋ a ɪ θ ɔ t ɐ k æ t	47	question	common_voice_en_39604205.mp3	test	true	5.184	d69e1833fa2a	false	unknown	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0.076923	0.014085	fite has said that she found the character unappealing i thought a cat	1	0	0	13	1	71	0.153846	0.042254	2	0	0	3	0.153846	0.028169	2	0	0	2	hite has said that she found the character unappealing i thought a cat	null	null	null	3.189228	3.492856	4.002994	-21.504187	0.819059	0.066	0	false	null	null	null	null	null	null	null	null	null	-23.73422
en-US_0010	en-US	main	cv17	17.0	She is the first Cypriot skater to ever qualify for the World Championships.	4.672	32,000	>=4.5	common_voice_en_32233084.mp3	dev	bafe6f99fffb	male	cv	teens	United States English	1	0.978596	0.546	1.182	0	-17.54954	0.980762	2.538986	3.325487	2.900318	What is it you want to research?	what is it you want to research	7	32	6-9	w ʌ t ɪ z ɪ t j u w ɔ n t t ə ɹ ᵻ s ɜ t ʃ	21	question	common_voice_en_18335001.mp3	test	true	1.824	e8d017393701	false	male		true	3.712	He then studied law with attorney John Colby of Washington.	general	[]	false	false	false	3.5-4.5	whisper	0	0	what is it you want to research	0	0	0	7	0	31	0.142857	0.096774	1	0	0	3	0	0	0	0	0	0	what is it you want to research	0.922902	0.790213	0.631114	2.834769	3.152327	3.878407	-10.852124	0.963816	0.066	0	false	2.971226	3.309845	3.858487	0	-33.522743	0.98222	0.066	0	false	-17.54954
en-US_0011	en-US	main	cv17	17.0	He also served twice as commandant of the New York Naval Shipyard.	3.872	48,000	3.7-4.5	common_voice_en_19155752.mp3	train	050b7c721443	female	cv	unknown	United States English	2	0.974174	0.802	1.954	0	-11.181969	0.990616	3.073642	3.539399	3.666549	What are you fishing after?	what are you fishing after	5	27	3-5	w ʌ t ɑ ɹ j u f ɪ ʃ ɪ ŋ æ f t ɚ	16	question	common_voice_en_39574195.mp3	test	true	1.536	e982b8b2ea66	false	unknown		true	3.232	If approved, without amendment, it is sent to the governor.	general	[]	false	false	false	<3.5	whisper	0	0	what are you fishing after	0	0	0	5	0	26	0	0	0	0	0	0	0	0	0	0	0	0	what are you fishing after	0.966624	0.631132	0.685298	2.182635	3.152119	2.431277	-12.952305	0.977865	0.034	0	false	2.594546	3.455631	2.799621	0	-13.748486	0.98948	0.034	0	false	-11.181969
en-US_0012	en-US	main	cv17	17.0	This makes it a safe way to try out software.	3.024	32,000	<3.7	common_voice_en_24683276.mp3	dev	65c7cd68a206	male	cv	fourties	United States English	1	0.988757	0.034	0	0	-36.858806	0.992431	2.935849	3.41905	3.705369	In Ireland, the butter is always salted, why would anyone want it any other way?	in ireland the butter is always salted why would anyone want it any other way	15	80	13-16	ɪ n a ɪ ɚ l ə n d ð ə b ʌ ɾ ɚ ɹ ɪ z ɔ l w e ɪ z s ɔ l t ᵻ d w a ɪ w ʊ d ɛ n ɪ w ʌ n w ɔ n t ɪ ɾ ɛ n i ʌ ð ɚ w e ɪ	57	question	common_voice_en_17297557.mp3	test	true	4.288	f8ad75855980	false	male		true	3.296	This is as directed by statute.	general	[]	false	false	false	<3.5	whisper	0	0	in ireland the butter is always salted why would anyone want it any other way	0	0	0	15	0	77	0.133333	0.012987	1	0	1	1	0	0	0	0	0	0	in ireland the butter is always salted why would anyone want it any other way	0.966364	0.625325	0.619354	2.875675	3.254207	3.759149	-23.155916	0.931437	0.034	0	false	2.674289	3.282163	3.315789	0	-38.540764	0.989684	0.034	0	false	-36.858806
en-US_0013	en-US	main	cv17	17.0	Mercer is primarily a writer focused on the gothic scene and its music.	4.096	32,000	3.7-4.5	common_voice_en_37228822.mp3	dev	282887ea4684	unknown	unknown	unknown	United States English	1	0.975586	1.09	0.602	0	-29.020849	0.995882	2.601677	3.02743	3.486536	A little bright-eyed terrier, you know, with oh, such long curly brown hair!	a little bright eyed terrier you know with oh such long curly brown hair	13	76	13-16	ɐ l ɪ ɾ ə l b ɹ a ɪ t a ɪ d t ɛ ɹ i ɚ j u n o ʊ w ɪ ð o ʊ s ʌ t ʃ l ɔ ŋ k ɜ l i b ɹ a ʊ n h ɛ ɹ	48	exclamation	common_voice_en_18489793.mp3	test	true	5.088	0157c27ca7de	false	male		true	3.584	The fruit has high contents of anthocyanins and ellagic acid.	general	[]	false	false	false	3.5-4.5	whisper	0.071429	0.041667	a little bright eyed terrier you know with such long curly brown hair	0	1	0	14	3	72	0.214286	0.125	2	1	0	9	0.071429	0.013889	1	0	0	1	a little bright eyed terrier you know with uh such long curly brown hair	0.981966	0.764208	0.699252	2.962023	3.223324	4.083224	-37.293617	0.916274	0.034	0	false	2.878141	3.295368	3.709751	0	-25.700363	0.981585	0.066	0	false	-29.020849
en-US_0014	en-US	main	cv17	17.0	But very soon it began to seem less of a game.	2.624	48,000	<3.7	common_voice_en_18634940.mp3	train	69a0609f88e1	female	cv	thirties	United States English	2	0.96189	0.674	0.594	0	-5.492326	0.958013	2.595702	3.470266	2.888525	What have you done?	what have you done	4	19	3-5	w ʌ t h æ v j u d ʌ n	11	question	common_voice_en_17353713.mp3	test	true	1.024	02a3d7c585f8	false	male		true	2.432	The plane descends steadily during landing.	general	[]	false	false	false	<3.5	whisper	0	0	what have you done	0	0	0	4	0	18	0	0	0	0	0	0	0	0	0	0	0	0	what have you done	0.926317	0.501834	0.504156	3.21899	3.490665	4.046463	-35.209488	0.935547	0.066	0	false	2.438268	3.057701	3.119402	0	-6.278144	0.972862	0.066	0	false	-5.492326
en-US_0015	en-US	main	cv17	17.0	Females give birth to seven to nine genetically identical offspring.	3.84	48,000	3.7-4.5	common_voice_en_18968664.mp3	train	a69b78a5b405	male	cv	thirties	United States English	1	0.973958	0.578	0.314	0	-40.925494	0.971981	2.925961	3.333389	3.696289	Long live the king!	long live the king	4	19	3-5	l ɔ ŋ l a ɪ v ð ə k ɪ ŋ	12	exclamation	common_voice_en_255533.mp3	test	true	1.472	052a589e28a2	false	male		true	4.16	Picigin must be played on a sandy beach in shallow water.	general	[]	false	false	false	3.5-4.5	whisper	0	0	long live the king	0	0	0	4	0	18	0.25	0.166667	1	0	0	3	0	0	0	0	0	0	long live the king	0.889284	0.585	0.507839	2.737431	3.322673	3.373593	-42.322449	0.976902	0.034	0	false	3.058085	3.433255	3.848996	0	-27.527378	0.944712	0.034	0	false	-40.925494
en-US_0016	en-US	main	cv17	17.0	Her debut film was "Manko Bandh".	4.16	32,000	3.7-4.5	common_voice_en_32412792.mp3	test	f13d09cfee97	female	cv	twenties	United States English	2	0.975962	0.13	0.67	0	-22.629246	0.973304	1.84842	3.182987	1.681999	Did you see Uncle Jake with any cash last night?	did you see uncle jake with any cash last night	10	48	10-12	d ɪ d j u s i ʌ ŋ k ə l d ʒ e ɪ k w ɪ ð ɛ n i k æ ʃ l æ s t n a ɪ t	34	question	common_voice_en_520490.mp3	test	true	2.48	05b87f9ab35e	false	male		true	5.44	Both the genus and the species were first described and published in Amer.	general	[]	false	false	false	>=4.5	whisper	0	0	did you see uncle jake with any cash last night	0	0	0	10	0	47	1	0.617021	10	0	0	29	0	0	0	0	0	0	did you see uncle jake with any cash last night	0.89457	0.562785	0.522283	2.201672	3.08619	2.853118	-17.631891	0.98629	0.034	0	false	2.258601	3.36882	2.407367	0	-15.649356	0.957721	0.066	0	false	-22.629246
en-US_0017	en-US	main	cv17	17.0	When the census confirms minority status, a meeting must be widely advertised.	4.64	48,000	>=4.5	common_voice_en_19777370.mp3	train	78a29cfb4cde	female	cv	fifties	United States English	2	0.978448	0.642	0.482	0	-23.825519	0.989004	3.012782	3.431853	3.736341	Where is that annoying sound coming from?	where is that annoying sound coming from	7	41	6-9	w ɛ ɹ ɪ z ð æ t ɐ n ɔ ɪ ɪ ŋ s a ʊ n d k ʌ m ɪ ŋ f ɹ ʌ m	28	question	common_voice_en_21290960.mp3	test	true	2.08	1107b35e0a27	false	male		true	3.36	It flows into the Garonne in Layrac, near Agen.	general	[]	false	false	false	<3.5	whisper	0	0	where is that annoying sound coming from	0	0	0	7	0	40	0	0	0	0	0	0	0	0	0	0	0	0	where is that annoying sound coming from	0.972411	0.664151	0.693102	2.003786	2.858675	2.551895	-9.052594	0.968269	0.066	0	false	2.942669	3.306717	3.834557	0	-32.898056	0.989881	0.034	0	false	-23.825519
en-US_0018	en-US	main	cv17	17.0	Kalchirburan is the nearest rural locality.	3.148	32,000	<3.7	common_voice_en_32836549.mp3	train	ff96183ed965	female	cv	fifties	Mid-Atlantic United States English,Philadelphia, Pennsylvania, United States English,United States English,Philadelphia Style United States English	2	0.984117	0.898	0	0	-16.173206	0.989485	2.002495	2.861819	2.475351	Oh no, my bunny broke out of its cage!	oh no my bunny broke out of its cage	9	38	6-9	o ʊ n o ʊ m a ɪ b ʌ n i b ɹ o ʊ k a ʊ ɾ ə v ɪ t s k e ɪ d ʒ	30	exclamation	common_voice_en_17679459.mp3	test	true	2.656	19ee3fd05db4	false	female		true	2.528	Records from Thailand are considered erroneous.	general	[]	false	false	false	<3.5	whisper	0	0	oh no my bunny broke out of its cage	0	0	0	9	0	36	0	0	0	0	0	0	0	0	0	0	0	0	oh no my bunny broke out of its cage	0.904862	0.588686	0.477517	2.863241	3.488923	3.445935	-29.829964	0.926205	0.066	0.13	false	2.482246	3.169459	3.168287	0	-12.084658	0.973892	0.066	0	false	-16.173206
en-US_0019	en-US	main	cv17	17.0	The second version also appears by itself as the B-side of the single.	4.256	32,000	3.7-4.5	common_voice_en_27368447.mp3	dev	3ffabdc32467	unknown	unknown	unknown	United States English	1	0.976504	0.898	1.066	0	-13.100279	0.993912	2.778761	3.486217	3.177239	Davison's autobiography, titled Is There Life Outside the Box?	davison's autobiography titled is there life outside the box	9	62	6-9	d æ v ɪ s ə n z ɔ ɾ o ʊ b a ɪ ɑ ɡ ɹ ə f i t a ɪ ɾ ə l d ɪ z ð ɛ ɹ l a ɪ f a ʊ t s a ɪ d ð ə b ɑ k s	50	question	common_voice_en_37042032.mp3	test	true	4.224	1a021494b7d5	false	unknown		true	3.232	"Domicile" here is a term with a technical meaning.	general	[]	false	false	false	<3.5	whisper	0	0	davison's autobiography titled is there life outside the box	0	0	0	9	0	60	0	0	0	0	0	0	0	0	0	0	0	0	davison's autobiography titled is there life outside the box	0.975183	0.760655	0.760929	3.217369	3.474461	4.107182	-18.958944	0.945549	0.034	0	false	3.102257	3.525472	3.759695	0	-17.847485	0.979579	0.066	0	false	-13.100279
en-US_0020	en-US	main	cv17	17.0	In the past Sambalpur has been a great centre of diamond trade.	3.808	32,000	3.7-4.5	common_voice_en_27003233.mp3	train	a6c35f97ef34	female	cv	thirties	United States English	2	0.973739	0.77	0.706	0	-16.537241	0.971519	2.959305	3.4208	3.702909	Don’t you understand me?	don t you understand me	4	24	3-5	d o ʊ n t j u ʌ n d ɚ s t æ n d m i	18	question	common_voice_en_18066799.mp3	test	true	1.568	042bdb6c2640	false	male		true	4.512	Executive Producer: Frank Chackler.	general	[]	false	false	false	>=4.5	whisper	0.4	0.043478	don't you understand me	1	1	0	5	1	23	0.4	0.043478	1	1	0	1	0.4	0.043478	1	1	0	1	don't you understand me	0.818002	0.407948	0.266105	2.432422	2.897591	3.43111	-2.580077	0.957908	0.066	0	false	3.27227	3.55905	4.112248	0	-16.37989	0.963209	0.066	0	false	-16.537241
en-US_0021	en-US	main	cv17	17.0	She was the daughter of Samuel S. Howland.	2.656	48,000	<3.7	common_voice_en_19706872.mp3	dev	989f9a2389aa	male	cv	fifties	United States English	1	0.962349	0.674	0.394	0	-13.868684	0.968088	2.948862	3.490692	3.53108	Have you been to Japan?	have you been to japan	5	23	3-5	h æ v j u b ɪ n t ə d ʒ ə p æ n	16	question	common_voice_en_17902458.mp3	test	true	1.792	0984a33b3e7c	false	male		true	3.616	He was offered a 'one year tour' playing with Alice.	general	[]	false	false	false	3.5-4.5	whisper	0	0	have you been to japan	0	0	0	5	0	22	0	0	0	0	0	0	0	0	0	0	0	0	have you been to japan	0.875886	0.445877	0.470812	3.280095	3.523353	4.12967	-26.573402	0.96317	0.066	0	false	2.531022	3.097374	3.312543	0	-26.907461	0.927544	0.13	0	false	-13.868684
en-US_0022	en-US	main	cv17	17.0	There's really no reason; it just rolled out that way.	2.592	48,000	<3.7	common_voice_en_19152839.mp3	test	8ae556d0f92d	male	cv	thirties	United States English	1	0.96142	0.834	0.37	0	-17.729761	1	2.1999	2.917144	2.940838	How was the food there?	how was the food there	5	23	3-5	h a ʊ w ʌ z ð ə f u d ð ɛ ɹ	14	question	common_voice_en_18361112.mp3	test	true	1.248	214de5e0a04c	false	male	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0	0	how was the food there	0	0	0	5	0	22	0	0	0	0	0	0	0	0	0	0	0	0	how was the food there	null	null	null	2.980929	3.450271	3.616319	-27.643538	0.947115	0.066	0	false	null	null	null	null	null	null	null	null	null	-17.729761
en-US_0023	en-US	main	cv17	17.0	Every day hundreds of happy adventurers land on its shores.	3.584	32,000	<3.7	common_voice_en_36438796.mp3	dev	dcc1a2cbde9a	unknown	unknown	unknown	United States English,Californian	1	0.972098	1.442	0.834	0	-29.258502	0.991205	3.068618	3.495868	3.716138	Don't you think so?	don't you think so	4	19	3-5	d o ʊ n t j u θ ɪ ŋ k s o ʊ	14	question	common_voice_en_18175978.mp3	test	true	1.12	2f7575450395	false	male		true	2.72	Lord Gravelton was blowing up the waiters.	general	[]	false	false	false	<3.5	whisper	0	0	don't you think so	0	0	0	4	0	18	0	0	0	0	0	0	0	0	0	0	0	0	don't you think so	0.951836	0.680549	0.723122	2.857996	3.298347	3.690094	-10.946531	0.969643	0.034	0	false	2.594281	3.209671	3.344624	0	-18.361145	0.975735	0.066	0	false	-29.258502
en-US_0024	en-US	main	cv17	17.0	Quackwatch's information is relevant to both consumers and medical professionals.	4.224	32,000	3.7-4.5	common_voice_en_37515621.mp3	test	9cc100a69e51	unknown	unknown	unknown	United States English	1	0.976326	0.898	0.378	0	-27.847392	1	2.563356	3.077598	3.392469	What shall I do?	what shall i do	4	16	3-5	w ʌ t ʃ æ l a ɪ d u	10	question	common_voice_en_17291279.mp3	test	true	1.248	3f1cf724d1ec	false	female	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0	0	what shall i do	0	0	0	4	0	15	0	0	0	0	0	0	0	0	0	0	0	0	what shall i do	null	null	null	2.671536	3.114202	3.561726	-25.070047	0.972756	0.034	0	false	null	null	null	null	null	null	null	null	null	-27.847392
en-US_0025	en-US	main	cv17	17.0	A little bright-eyed terrier, you know, with oh, such long curly brown hair!	5.088	48,000	>=4.5	common_voice_en_18489793.mp3	test	0157c27ca7de	male	cv	thirties	United States English	1	0.909591	1.538	0.698	0	-29.280145	1	2.962023	3.223324	4.083224	Will you marry me?	will you marry me	4	18	3-5	w ɪ l j u m æ ɹ i m i	11	question	common_voice_en_17452908.mp3	test	true	1.248	4b7360b2e1f5	false	male	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0	0	will you marry me	0	0	0	4	0	17	0	0	0	0	0	0	0	0	0	0	0	0	will you marry me	null	null	null	2.82849	3.136935	3.929303	-38.991211	0.947115	0.066	0	false	null	null	null	null	null	null	null	null	null	-29.280145
en-US_0026	en-US	main	cv17	17.0	The artist made a mistake with the boundaries between Peru and Ecuador.	5.472	48,000	>=4.5	common_voice_en_19954083.mp3	test	277c4e5f5fd1	male	cv	thirties	United States English	1	0.981725	0.93	0.418	0.000003	-16.617258	1	2.548212	3.029835	3.562252	Can you file these files, please?	can you file these files please	6	33	6-9	k æ n j u f a ɪ l ð i z f a ɪ l z p l i z	21	question	common_voice_en_18041350.mp3	test	true	1.848	4ed7e3dd3845	false	male	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0.666667	0.419355	kenny baldy's files please	2	2	0	6	13	31	1	0.709677	5	1	0	22	0.166667	0.129032	1	0	0	4	can you draw these files please	null	null	null	1.833666	2.955625	1.960436	-19.771666	0.964286	0.066	0	false	null	null	null	null	null	null	null	null	null	-16.617258
en-US_0027	en-US	main	cv17	17.0	The locomotives were built in two series by various manufacturers.	4.608	32,000	>=4.5	common_voice_en_32301670.mp3	train	50a6b43fb8b5	male	cv	thirties	United States English	1	0.978299	0.898	0.714	0	-22.933738	0.991198	3.143015	3.432392	3.966089	Do you ever doodle things on a piece of paper when you're bored?	do you ever doodle things on a piece of paper when you're bored	13	64	13-16	d u j u ɛ v ɚ d u d ə l θ ɪ ŋ z ɔ n ɐ p i s ʌ v p e ɪ p ɚ w ɛ n j ʊ ɹ b o ɹ d	39	question	common_voice_en_18024531.mp3	test	true	4.896	54b513175262	false	male		true	3.008	She resolves to continue living with her head held high.	general	[]	false	false	false	<3.5	whisper	0.307692	0.079365	do you ever do the things on a piece of paper when you are bored	2	0	2	13	5	63	0.384615	0.111111	3	0	2	7	0.153846	0.031746	1	0	1	2	do you ever doodle things on a piece of paper when you are bored	0.968799	0.713314	0.68693	2.945757	3.477724	3.497294	-15.27436	0.836193	0.802	0	false	2.845205	3.551625	3.217748	0	-26.915743	0.988697	0.034	0	false	-22.933738
en-US_0028	en-US	main	cv17	17.0	in parliament Antiquity- Even the women knew how to be silent	4.192	48,000	3.7-4.5	common_voice_en_677891.mp3	dev	4372bfbfa7cb	male	cv	thirties	United States English	1	0.875	1.922	1.09	0	-22.749602	1	3.212409	3.472029	4.076532	What do you want of me?	what do you want of me	6	23	6-9	w ʌ t d u j u w ɔ n t ʌ v m i	15	question	common_voice_en_17895168.mp3	test	true	1.632	5ae2ccf9bfd9	false	female	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0	0	what do you want of me	0	0	0	6	0	22	0.333333	0.090909	2	0	0	2	0	0	0	0	0	0	what do you want of me	null	null	null	2.387131	3.353697	2.630078	-30.743446	0.959559	0.066	0	false	null	null	null	null	null	null	null	null	null	-22.749602
en-US_0029	en-US	main	cv17	17.0	It is named after the Tokaido route of Japan, used for centuries.	4.8	48,000	>=4.5	common_voice_en_20289513.mp3	train	1f149cd636e4	female	cv	thirties	United States English	2	0.979167	0.93	0.73	0	-24.947972	0.974551	3.313877	3.567673	4.053917	What more do you want?	what more do you want	5	22	3-5	w ʌ t m o ɹ d u j u w ɔ n t	14	question	common_voice_en_17562784.mp3	test	true	1.328	5efda2ff7f07	false	male		true	2.208	This cannot be a coincidence.	general	[]	false	false	false	<3.5	whisper	0	0	what more do you want	0	0	0	5	0	21	0	0	0	0	0	0	0	0	0	0	0	0	what more do you want	0.894487	0.588317	0.487194	2.870913	3.26786	3.826635	-38.872452	0.902108	0.13	0	false	3.194454	3.677321	3.677492	0	-24.150719	0.984601	0.034	0	false	-24.947972
en-US_0030	en-US	main	cv17	17.0	Not to break is better than to mend.	2.784	48,000	<3.7	common_voice_en_17884083.mp3	test	d0f5f9e233c4	male	cv	teens	United States English	1	0.96408	0.578	0.506	0	-27.168274	1	2.437546	2.911243	3.443674	Why do people even eat peanut butter and jelly?	why do people even eat peanut butter and jelly	9	47	6-9	w a ɪ d u p i p ə l i v ə n i t p i n ʌ t b ʌ ɾ ɚ æ n d d ʒ ɛ l i	33	question	common_voice_en_17551784.mp3	test	true	2.624	5f7a6e8ab8b6	false	male	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0	0	why do people even eat peanut butter and jelly	0	0	0	9	0	46	0.333333	0.086957	2	0	1	4	0	0	0	0	0	0	why do people even eat peanut butter and jelly	null	null	null	3.29817	3.521676	4.125645	-23.014097	0.987043	0.034	0	false	null	null	null	null	null	null	null	null	null	-27.168274
en-US_0031	en-US	main	cv17	17.0	Regarding your request, I have decided to heed your warnings.	3.808	48,000	3.7-4.5	common_voice_en_17286891.mp3	dev	2fb94f08efe1	male	cv	thirties	United States English	1	0.921218	0.866	2.602	0	-27.674745	1	3.135108	3.58528	3.747355	Is Anybody Out There?	is anybody out there	4	21	3-5	ɪ z ɛ n ɪ b ɑ d i a ʊ t ð ɛ ɹ	15	question	common_voice_en_39569399.mp3	test	true	1.248	647ff86a37d9	false	unknown	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0	0	is anybody out there	0	0	0	4	0	20	0.25	0.1	1	0	0	2	0	0	0	0	0	0	is anybody out there	null	null	null	2.701773	3.089757	3.680171	-38.076645	0.972756	0.034	0	false	null	null	null	null	null	null	null	null	null	-27.674745
en-US_0032	en-US	main	cv17	17.0	The next day he was raised to the peerage as Baron Dunfermline.	4.416	32,000	3.7-4.5	common_voice_en_33896600.mp3	dev	4f4c4f357a0c	unknown	unknown	unknown	United States English	1	0.977355	0.45	0.67	0	-21.976941	0.989429	2.158352	2.736756	3.168841	Security, kick this guy out!	security kick this guy out	5	28	3-5	s ᵻ k j ʊ ɹ ɹ ᵻ ɾ i k ɪ k ð ɪ s ɡ a ɪ a ʊ t	22	exclamation	common_voice_en_17881800.mp3	test	true	2.656	79feeb11c5db	false	male		true	3.68	The Great Forester River forms part of the southern boundary.	general	[]	false	false	false	3.5-4.5	whisper	0	0	security kick this guy out	0	0	0	5	0	26	0.8	0.307692	3	1	0	8	0.2	0.115385	1	0	0	3	security check this guy out	0.955139	0.774651	0.665669	2.698252	3.022385	3.799402	-24.081913	0.768825	0.098	0	false	1.826442	2.467617	2.798341	0	-19.578909	0.990761	0.034	0	false	-21.976941
en-US_0033	en-US	main	cv17	17.0	The music also incorporated more elements of electronic loops, synthesizers and studio effects.	5.224	32,000	>=4.5	common_voice_en_25269720.mp3	train	b4780ed0fb71	male	cv	twenties	United States English	1	0.990429	0.226	0	0	-24.914667	0.984102	3.225629	3.561001	3.918004	Who does one tell first?	who does one tell first	5	24	3-5	h u d ʌ z w ʌ n t ɛ l f ɜ s t	15	question	common_voice_en_39588526.mp3	test	true	1.632	838f4c3563f8	false	unknown		true	2.912	But you have the advantage of numbers against him.	general	[]	false	false	false	<3.5	whisper	0	0	who does one tell first	0	0	0	5	0	23	0.2	0.043478	1	0	0	1	0	0	0	0	0	0	who does one tell first	0.939869	0.567175	0.581383	2.005725	2.861152	2.657456	-35.112659	0.979167	0.034	0	false	3.237407	3.444764	4.130643	0	-30.88302	0.977335	0.066	0	false	-24.914667
en-US_0034	en-US	main	cv17	17.0	If I had the choice between honey and jam, I would choose the latter.	4.768	48,000	>=4.5	common_voice_en_17668574.mp3	test	24f68ef5c02b	male	cv	fifties	United States English	1	0.958054	0.674	0.202	0	-19.988446	1	3.19899	3.647119	3.769219	Great, can you show them to me?	great can you show them to me	7	31	6-9	ɡ ɹ e ɪ t k æ n j u ʃ o ʊ ð ɛ m t ə m i	20	question	common_voice_en_17255702.mp3	test	true	2.336	a392ee9796e6	false	male	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0	0	great can you show them to me	0	0	0	7	0	29	0	0	0	0	0	0	0	0	0	0	0	0	great can you show them to me	null	null	null	2.938584	3.521363	3.451845	-22.091759	0.778253	0.066	0	false	null	null	null	null	null	null	null	null	null	-19.988446
en-US_0035	en-US	main	cv17	17.0	In these depressions, snow persisted through summer months, and becomes glacial ice.	5.152	48,000	>=4.5	common_voice_en_22749494.mp3	train	ba55291b3ae5	female	cv	teens	United States English	2	0.98059	0.674	0.97	0	-21.929374	0.953478	2.44471	3.344823	2.694741	Who told you that?	who told you that	4	18	3-5	h u t o ʊ l d j u ð æ t	12	question	common_voice_en_17904824.mp3	test	true	1.088	a66c026d2c67	false	male		true	2.272	The idea was his sister Margaret's.	general	[]	false	false	false	<3.5	whisper	0.75	0.647059	what are you doing	3	0	0	4	11	17	0.75	0.529412	3	0	0	9	0.25	0.294118	1	0	0	5	who taught you that	0.937017	0.626008	0.607628	2.305846	3.305147	2.462127	-33.985172	0.939338	0.066	0	false	3.044119	3.568697	3.491794	0	-19.960049	0.985035	0.034	0	false	-21.929374
en-US_0036	en-US	main	cv17	17.0	Oh my dear paws!	2.08	48,000	<3.7	common_voice_en_18450447.mp3	dev	7248f78f3969	female	cv	twenties	United States English	2	0.951923	0.482	1.954	0	-29.064963	1	1.901716	2.489873	2.436201	Where's that airlines bag?	where's that airlines bag	4	26	3-5	w ɛ ɹ z ð æ t ɛ ɹ l a ɪ n z b æ ɡ	17	question	common_voice_en_199114.mp3	dev	true	2.016	2229dc384d20	false	male	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0.25	0.04	where's that airline's bag	1	0	0	4	1	25	0.25	0.04	1	0	0	1	0	0	0	0	0	0	where's that airlines bag	null	null	null	2.209554	3.424502	2.104681	-27.609467	0.951389	0.098	0	false	null	null	null	null	null	null	null	null	null	-29.064963
en-US_0037	en-US	main	cv17	17.0	Civilized Creatures in "Spore" can also be taught to speak Simlish.	4.512	48,000	>=4.5	common_voice_en_19768149.mp3	train	c34cad469fe2	female	cv	twenties	United States English	2	0.977837	0.578	0.626	0	-17.749605	0.992823	3.359248	3.605055	4.117298	Can you help me find Preaching to the Perverted?	can you help me find preaching to the perverted	9	48	6-9	k æ n j u h ɛ l p m i f a ɪ n d p ɹ i t ʃ ɪ ŋ t ə ð ə p ɚ v ɜ ɾ ᵻ d	34	question	common_voice_en_46084.mp3	dev	true	2.912	4dc00be70c01	false	male		true	3.36	There is rarely any space for them to move in any case.	general	[]	false	false	false	<3.5	whisper	0	0	can you help me find preaching to the perverted	0	0	0	9	0	47	0	0	0	0	0	0	0	0	0	0	0	0	can you help me find preaching to the perverted	0.969766	0.647663	0.638499	2.865529	3.548334	3.213743	-18.928125	0.988324	0.034	0	false	3.089671	3.429625	3.877641	0	-20.377884	0.980357	0.066	0	false	-17.749605
en-US_0038	en-US	main	cv17	17.0	Hence it is called the "sacrament of perfection" or the "queen of sacraments".	4.576	48,000	>=4.5	common_voice_en_20302269.mp3	train	bafb31f40d7d	female	cv	twenties	United States English	2	0.978147	0.898	0.698	0	-11.878746	0.956636	2.856814	3.433448	3.438774	God save the Queen!	god save the queen	4	19	3-5	ɡ ɑ d s e ɪ v ð ə k w i n	13	exclamation	common_voice_en_20535542.mp3	dev	true	2.848	52e15d3d8769	false	male		true	3.264	This process is lengthy, and can be difficult and costly.	general	[]	false	false	false	<3.5	whisper	0.25	0.055556	god saved the queen	1	0	0	4	1	18	0.25	0.111111	1	0	0	2	0.25	0.333333	0	0	1	6	god save the queen hello	0.900632	0.580402	0.551788	2.603131	3.400555	2.966172	-31.685375	0.615871	0.13	0	true	3.08938	3.56508	3.706682	0	-20.214127	0.979779	0.066	0	false	-11.878746
en-US_0039	en-US	main	cv17	17.0	It has also been featured on "Deadly Women".	2.752	32,000	<3.7	common_voice_en_39588095.mp3	test	b64eed6c63f5	unknown	unknown	unknown	United States English,portuguese	1	0.963663	0.61	1.022	0	-19.757977	1	2.766685	3.164752	3.746784	Tell me frankly, Jeeves, are you in pretty good shape mentally?	tell me frankly jeeves are you in pretty good shape mentally	11	63	10-12	t ɛ l m i f ɹ æ ŋ k l i d ʒ i v z ɑ ɹ j u ɪ n p ɹ ɪ ɾ i ɡ ʊ d ʃ e ɪ p m ɛ n t ə l i	42	question	common_voice_en_18592570.mp3	dev	true	4.544	5e6e088562d6	false	male	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0	0	tell me frankly jeeves are you in pretty good shape mentally	0	0	0	11	0	60	0.090909	0.05	1	0	0	3	0	0	0	0	0	0	tell me frankly jeeves are you in pretty good shape mentally	null	null	null	2.948149	3.313652	3.786977	-19.05842	0.793574	0.034	0	false	null	null	null	null	null	null	null	null	null	-19.757977
en-US_0040	en-US	main	cv17	17.0	The novel contains autobiographical elements.	3.008	32,000	<3.7	common_voice_en_39577803.mp3	test	1e258a6ba23b	unknown	unknown	unknown	United States English	1	0.966755	1.122	0.686	0	-23.423861	1	2.494156	3.060555	3.206855	They'd never let you go!	they'd never let you go	5	24	3-5	ð e ɪ d n ɛ v ɚ l ɛ t j u ɡ o ʊ	16	exclamation	common_voice_en_56143.mp3	dev	true	1.632	5f2b5e7619e9	false	male	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0	0	they'd never let you go	0	0	0	5	0	23	0	0	0	0	0	0	0	0	0	0	0	0	they'd never let you go	null	null	null	3.319495	3.591023	4.07898	-49.535999	0.959559	0.066	0	true	null	null	null	null	null	null	null	null	null	-23.423861
en-US_0041	en-US	main	cv17	17.0	Homeotic mutations work by changing segment identity during development.	4.256	48,000	3.7-4.5	common_voice_en_20994303.mp3	train	8999c6389b4b	female	cv	twenties	United States English	2	0.976504	0.866	0.738	0	-19.303849	0.99022	2.950521	3.506876	3.561992	And something else changed, I started having fun!	and something else changed i started having fun	8	49	6-9	æ n d s ʌ m θ ɪ ŋ ɛ l s t ʃ e ɪ n d ʒ d a ɪ s t ɑ ɹ ɾ ᵻ d h æ v ɪ ŋ f ʌ n	37	exclamation	common_voice_en_648613.mp3	dev	true	3.296	6e06ba34f7ef	false	male		true	3.072	The former crop is currently growing wild in the area.	general	[]	false	false	false	<3.5	whisper	0.125	0.085106	and something else had changed i started having fun	0	0	1	8	4	47	0.125	0.085106	0	0	1	4	0.125	0.085106	0	0	1	4	and something else had changed i started having fun	0.950492	0.633399	0.605052	3.006476	3.414288	3.770574	-9.948061	0.784587	0.098	0	false	3.443528	3.682439	4.155765	0	-27.328823	0.988932	0.034	0	false	-19.303849
en-US_0042	en-US	main	cv17	17.0	The badge on the front grille was an option at first.	3.232	48,000	<3.7	common_voice_en_19933816.mp3	test	1e61bea07830	male	cv	thirties	United States English	1	0.969059	0.546	0.69	0	-33.549904	1	1.875051	2.584725	2.276478	On what point?	on what point	3	14	3-5	ɔ n w ʌ t p ɔ ɪ n t	10	question	common_voice_en_18450446.mp3	dev	true	1.088	7248f78f3969	false	female	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0	0	on what point	0	0	0	3	0	13	0	0	0	0	0	0	0	0	0	0	0	0	on what point	null	null	null	2.327169	2.788384	3.410649	-29.036163	0.96875	0.034	0	false	null	null	null	null	null	null	null	null	null	-33.549904
en-US_0043	en-US	main	cv17	17.0	They are light gray on top with a white rump and a white underside.	4.8	32,000	>=4.5	common_voice_en_39734854.mp3	test	dcc2fd101b13	female	cv	teens	United States English,England English	2	0.951667	1.058	0.578	0	-27.48483	1	2.715775	3.116514	3.686523	Oh my dear paws!	oh my dear paws	4	16	3-5	o ʊ m a ɪ d ɪ ɹ p ɔ z	11	exclamation	common_voice_en_18450447.mp3	dev	true	2.08	7248f78f3969	false	female	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0.25	0.133333	oh my dear paul	1	0	0	4	2	15	0.5	0.2	2	0	0	3	0.25	0.133333	1	0	0	2	oh my dear paul	null	null	null	1.901716	2.489873	2.436201	-31.827078	0.825962	0.066	0	false	null	null	null	null	null	null	null	null	null	-27.48483
en-US_0044	en-US	main	cv17	17.0	He became particularly interested in computer network security.	3.872	48,000	3.7-4.5	common_voice_en_22355203.mp3	train	b41887200545	female	cv	twenties	United States English	2	0.974174	0.994	0.61	0	-28.20536	0.958726	2.782073	3.307983	3.492986	Oh, I am not unhappy, cousin John!	oh i am not unhappy cousin john	7	34	6-9	o ʊ a ɪ ɐ m n ɑ t ʌ n h æ p i k ʌ z ə n d ʒ ɑ n	24	exclamation	common_voice_en_18565873.mp3	dev	true	3.968	84a278ccb9f1	false	female		true	2.08	This show is hosted by Jeremiah Burton.	general	[]	false	false	false	<3.5	whisper	0.285714	0.064516	oh i'm not unhappy cousin john	1	1	0	7	2	31	0.571429	0.16129	3	1	0	5	0.285714	0.064516	1	1	0	2	oh i'm not unhappy cousin john	0.878493	0.502081	0.468909	2.59176	3.04788	3.471715	-24.525997	0.634577	0.13	0	true	3.141061	3.446558	4.004931	0	-29.451754	0.983654	0.034	0	false	-28.20536
en-US_0045	en-US	main	cv17	17.0	There are three colleges and one pre-university.	3.04	32,000	<3.7	common_voice_en_39644682.mp3	test	83bd9a42966a	unknown	unknown	unknown	United States English	1	0.967105	1.41	1.266	0	-33.700422	1	2.656522	3.195215	3.432743	He has also produced two collections of gag cartoons, Haw!	he has also produced two collections of gag cartoons haw	10	58	10-12	h i h ɐ z ɔ l s o ʊ p ɹ ə d u s t t u k ə l ɛ k ʃ ə n z ʌ v ɡ æ ɡ k ɑ ɹ t u n z h ɔ	42	exclamation	common_voice_en_23699312.mp3	dev	true	4.48	8b42a3b5b6cc	false	male	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0.1	0.017857	he has also produced two collections of gag cartoons ha	1	0	0	10	1	56	0.3	0.125	1	2	0	7	0.1	0.017857	1	0	0	1	he has also produced two collections of gag cartoons ha	null	null	null	2.832515	3.169703	3.866852	-29.640574	0.985268	0.066	0	false	null	null	null	null	null	null	null	null	null	-33.700422
en-US_0046	en-US	main	cv17	17.0	Having more than five members is mostly uncommon in rock and pop music.	4.192	32,000	3.7-4.5	common_voice_en_24994091.mp3	train	d92fb8a107b5	male	cv	twenties	United States English	1	0.976145	0.802	1.082	0	-25.613158	0.978171	1.927995	3.089372	2.111912	'How charming is that?	'how charming is that	4	22	3-5	h a ʊ t ʃ ɑ ɹ m ɪ ŋ ɪ z ð æ t	15	question	common_voice_en_21831690.mp3	dev	true	1.856	93b3fe225ff5	false	male		true	2.08	Autumn frosts have slain July.	general	[]	false	false	false	<3.5	whisper	0.25	0.047619	how charming is that	1	0	0	4	1	21	0.25	0.047619	1	0	0	1	0.25	0.047619	1	0	0	1	how charming is that	0.96601	0.661981	0.542751	2.980165	3.430861	3.720738	-16.65192	0.96444	0.066	0	false	1.370619	2.219737	1.712345	0	-16.030069	0.983654	0.034	0	false	-25.613158
en-US_0047	en-US	main	cv17	17.0	Conservation of the island is managed by the Island Conservation Society.	5.12	48,000	>=4.5	common_voice_en_20489834.mp3	train	76a32d77b283	female	cv	teens	United States English	2	0.980469	0.706	0.61	0	-24.060816	0.994113	2.814175	2.998168	4.089564	Will it be stormy in Old Fort Saint Lucia right now?	will it be stormy in old fort saint lucia right now	11	52	10-12	w ɪ l ɪ t b i s t o ɹ m i ɪ n o ʊ l d f ɔ ɹ t s e ɪ n t l u ʃ ɚ ɹ a ɪ t n a ʊ	39	question	common_voice_en_649351.mp3	dev	true	2.672	95b934de06c8	false	female		true	3.552	The menacing creatures would often disappear at dawn.	general	[]	false	false	false	3.5-4.5	whisper	0.090909	0.058824	will it be stormy in old fort st lucia right now	1	0	0	11	3	51	0.727273	0.313726	7	1	0	16	0.090909	0.058824	1	0	0	3	will it be stormy in old fort st lucia right now	0.97582	0.706885	0.683649	2.273546	3.161778	2.698349	-41.24654	0.987275	0.034	0	false	2.665531	3.030161	3.657609	0	-23.948875	0.97241	0.098	0	false	-24.060816
en-US_0048	en-US	main	cv17	17.0	She taught at Xinjiang University prior to her detention.	3.584	32,000	<3.7	common_voice_en_33606184.mp3	dev	15ed34d59e1b	male	cv	twenties	United States English	1	0.972098	0.61	1.414	0	-18.952604	0.977535	3.220087	3.48735	4.057545	He had asked contestant Heather Cook, Are you male or female?	he had asked contestant heather cook are you male or female	11	61	10-12	h i h æ d æ s k t k ə n t ɛ s t ə n t h ɛ ð ɚ k ʊ k ɑ ɹ j u m e ɪ l ɔ ɹ f i m e ɪ l	42	question	common_voice_en_37513698.mp3	dev	true	4.48	ac4cada8536f	false	unknown		true	3.392	Kareem Daniel grew up on the South Side of Chicago.	general	[]	false	false	false	<3.5	whisper	0	0	he had asked contestant heather cook are you male or female	0	0	0	11	0	59	0.090909	0.016949	1	0	0	1	0	0	0	0	0	0	he had asked contestant heather cook are you male or female	0.93643	0.630941	0.67558	2.255075	3.283819	2.394567	-16.377684	0.898661	0.066	0	false	2.847219	3.329293	3.610357	0	-9.559978	0.989976	0.034	0	false	-18.952604
en-US_0049	en-US	main	cv17	17.0	In such cases the festival assumes less imposing dimensions.	5.088	32,000	>=4.5	common_voice_en_38257057.mp3	test	2071ecf8a422	unknown	unknown	unknown	United States English	1	0.980346	0.45	0.718	0	-13.427308	1	3.202348	3.569221	3.868888	The Smallest Ship that Ever Crossed the Atlantic!	the smallest ship that ever crossed the atlantic	8	49	6-9	ð ə s m ɔ l ɪ s t ʃ ɪ p ð æ t ɛ v ɚ k ɹ ɔ s t ð ɪ ɐ t l æ n t ɪ k	33	exclamation	common_voice_en_32704289.mp3	dev	true	2.88	c89c98e09ba2	false	male	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0	0	the smallest ship that ever crossed the atlantic	0	0	0	8	0	48	0.25	0.145833	2	0	0	7	0	0	0	0	0	0	the smallest ship that ever crossed the atlantic	null	null	null	1.740661	1.915435	3.570644	-88.208633	0.977083	0.066	0	true	null	null	null	null	null	null	null	null	null	-13.427308
en-US_0050	en-US	main	cv17	17.0	The tute bianches have had international variations of one sort or another.	5.088	32,000	>=4.5	common_voice_en_37177061.mp3	test	46e4741a2e22	unknown	unknown	unknown	United States English,Gay	1	0.954403	0.546	0.586	0	-28.73481	1	2.806154	3.173098	3.819374	Who will win the arm wrestling match?	who will win the arm wrestling match	7	37	6-9	h u w ɪ l w ɪ n ð ɪ ɑ ɹ m ɹ ɛ s ə l ɪ ŋ m æ t ʃ	24	question	common_voice_en_25363324.mp3	dev	true	2.912	d30e81efe3e1	false	male	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0	0	who will win the arm wrestling match	0	0	0	7	0	36	0.571429	0.25	3	1	0	9	0	0	0	0	0	0	who will win the arm wrestling match	null	null	null	2.19004	3.076192	2.818012	-23.822313	0.988324	0.034	0	false	null	null	null	null	null	null	null	null	null	-28.73481
en-US_0051	en-US	main	cv17	17.0	The highway is also receiving a new routing to bypass Greenville, Mississippi.	4.224	48,000	3.7-4.5	common_voice_en_19693860.mp3	train	43fbfe4348c6	male	cv	thirties	United States English	1	0.976326	1.218	0.85	0	-23.010368	0.985078	3.31069	3.595301	4.081866	Is there no sense of honor left in this country?	is there no sense of honor left in this country	10	48	10-12	ɪ z ð ɛ ɹ n o ʊ s ɛ n s ʌ v ɑ n ɚ l ɛ f t ɪ n ð ɪ s k ʌ n t ɹ i	32	question	common_voice_en_171532.mp3	dev	true	2.848	f10ab695ba41	false	male		true	3.424	The rest of the hornbill's plumage is a glossy dark-green and black.	general	[]	false	false	false	<3.5	whisper	0	0	is there no sense of honor left in this country	0	0	0	10	0	47	0.1	0.021277	1	0	0	1	0	0	0	0	0	0	is there no sense of honor left in this country	0.94721	0.75694	0.619765	3.143065	3.449167	3.983878	-33.17231	0.976826	0.066	0	false	3.074507	3.459491	3.849287	0	-25.739164	0.980724	0.066	0	false	-23.010368
en-US_0052	en-US	main	cv17	17.0	The village is the site of Utne Church.	3.008	48,000	<3.7	common_voice_en_20015291.mp3	test	e30120248813	male	cv	twenties	United States English	1	0.966755	1.09	1.354	0	-22.543293	1	2.244431	2.676498	3.440473	Just as I thought!	just as i thought	4	18	3-5	d ʒ ʌ s t æ z a ɪ θ ɔ t	12	exclamation	common_voice_en_18009689.mp3	dev	true	3.4	0332e7147236	false	male	Not supported with pagination yet	false	0		general	[]	false	false	false	n/a	whisper	0	0	just as i thought	0	0	0	4	0	17	0	0	0	0	0	0	0	0	0	0	0	0	just as i thought	null	null	null	1.971324	2.503967	3.068512	-22.669552	0.697059	0.034	0	true	null	null	null	null	null	null	null	null	null	-22.543293
en-US_0053	en-US	main	cv17	17.0	He appeased Hugh and Margaret by creating Hugh Earl of Gloucester.	3.552	32,000	<3.7	common_voice_en_37110327.mp3	dev	e51b0e6ce076	unknown	unknown	unknown	United States English,Northeast US	1	0.971847	0.642	0.766	0	-22.806758	0.981776	3.101471	3.451946	3.816428	What do you mean by that?	what do you mean by that	6	25	6-9	w ʌ t d u j u m i n b a ɪ ð æ t	16	question	common_voice_en_2462108.mp3	dev	true	1.12	05008b93902a	false	male		true	2.4	The album's songs are all in Swedish.	general	[]	false	false	false	<3.5	whisper	0	0	what do you mean by that	0	0	0	6	0	24	0.166667	0.125	0	1	0	3	0	0	0	0	0	0	what do you mean by that	0.906829	0.657306	0.649167	1.587437	2.13654	2.845278	-50.623898	0.941071	0.066	0	true	3.080231	3.481603	3.854984	0	-23.611328	0.985833	0.034	0	false	-22.806758

End of preview. Expand in Data Studio

Multilingual Speech Benchmark for Zero-Shot TTS

A voice-cloning and intelligibility benchmark for 8 language subsets, with 10,100 examples selected from Common Voice 17.0. Each example supplies a speaker reference and an independently selected target text, with human recordings as WER/CER and speaker-similarity anchors when the corresponding audio is available. Corpus WER and the WavLM-FT evaluation follow seed-tts-eval.

Version 3.1. Adds ru, kk through the same S1–S4 selection methodology, followed by Sidon restoration of both prompt_audio and sim_ref_audio. All six subsets already present in v3.0, including restored Kyrgyz, retain their packaged audio, rows and per-utterance measurements. The extension validator checks this against the preserved parent revision v3.0 at commit f1f614c12a14b658bbc3ad5776dbf6f2b87a59ff. Existing results for those subsets remain valid.

Pipeline: source repository; the release's exact sources are bundled in pipeline_source.tar.gz. manifest.json records the parent, new subset composition and selection settings. See LANGUAGES_ru_kk.md for the new-language recipe and RESTORATION.md for historical Kyrgyz restoration.

Quick start

from datasets import load_dataset

ds = load_dataset("nineninesix/multilingual-tts-benchmark", "ru", split="main", revision="v3.1")
row = ds[0]
row["prompt_audio"]   # condition a voice-cloning model on this recording
row["text"]           # synthesise this independently selected target text
row["gt_audio"]       # human reading used for the intelligibility anchor
row["sim_ref_audio"]  # separate recording of the reference speaker

wer = load_dataset("nineninesix/multilingual-tts-benchmark", "metrics-wer", split="train", revision="v3.1")
sim = load_dataset("nineninesix/multilingual-tts-benchmark", "metrics-sim", split="train", revision="v3.1")

The four metric configs remain metrics-wer, metrics-sim, metrics-sim-floor and metrics-quality; their artifacts include the new languages alongside the preserved measurements. Missing audio and unavailable optional ASR results are represented by flags and nulls, rather than invented measurements.

msbench-fetch --pack tts-bench-v3.1 --revision v3.1
msbench-wer --pack tts-bench-v3.1 --lang ru --audio synth --asr whisper \
    --synth-dir out/mymodel/ru --model-name mymodel
msbench-sim --pack tts-bench-v3.1 --lang ru --mode synth --encoder wavlm_ft \
    --synth-dir out/mymodel/ru
msbench-report --pack tts-bench-v3.1 --all

For Kazakh, use --lang kk --asr gigaam. Preprocess synthesis with msbench.audio.prepare(); feed stored benchmark audio directly to the encoders without trimming it again.

Subsets

subset	examples	speakers used	with GT audio	with second clip	phone units	diphones
en-US	1500	1162	1500/1500	909/1500	49	1489
es-ES	1500	722	1499/1500	1222/1500	37	808
es-MX	1500	929	1500/1500	1201/1500	36	785
nl-NL	1500	469	1500/1500	1319/1500	44	1083
pt-BR	1500	247	1500/1500	983/1500	43	1080
ky	700	181	694/700	599/700	32	741
ru	1500	1299	1500/1500	1383/1500	53	1803
kk	400	120	400/400	325/400	33	575

Full composition and candidate-pool coverage evidence: en-US, es-ES, es-MX, nl-NL, pt-BR, ky, ru, kk. The inventory is the one emitted by the documented eSpeak voice/tokenisation; it is not an independently established linguistic inventory. Russian uses native eSpeak phone units, including palatalised consonants and affricates, u" and ɪ^; explicit boundaries preserve these as atomic phones instead of inventing standalone ʲ, " or ^ units. These internal symbols are defined by the Russian eSpeak voice.

Human anchors — intelligibility

Primary backends are configured per language in msbench.languages. These anchors score real human readings of the target text, including reader and transcript errors as well as ASR errors. The headline is corpus WER, Σ(S+D+I)/Σ N_ref. Intervals use speaker-cluster bootstraps.

lang	ASR	n	speakers	WER corpus	95% CI	CER corpus	exact	catastrophic
en-US	whisper	1500	1012	0.0774	[0.0703, 0.0842]	0.0296	59.7%	1.4%
es-ES	whisper	1499	574	0.0452	[0.0397, 0.0512]	0.0154	73.0%	0.9%
es-MX	whisper	1500	639	0.0613	[0.0541, 0.0691]	0.0214	68.3%	1.3%
nl-NL	whisper	1500	236	0.0385	[0.0331, 0.0453]	0.0108	76.0%	0.5%
pt-BR	whisper	1500	229	0.0737	[0.0608, 0.0884]	0.0234	71.5%	3.5%
ky	gigaam	694	89	0.0998	[0.0773, 0.1311]	0.0293	66.4%	5.5%
ru	whisper	1500	440	0.0543	[0.0473, 0.0628]	0.0148	71.8%	1.1%
kk	gigaam	400	64	0.1179	[0.0930, 0.1417]	0.0263	57.8%	3.2%

All measured recognisers

lang	ASR	n	speakers	WER corpus	95% CI	CER corpus	exact	catastrophic
en-US	elevenlabs	1500	1012	0.0515	[0.0462, 0.0569]	0.0202	70.4%	1.0%
en-US	whisper	1500	1012	0.0774	[0.0703, 0.0842]	0.0296	59.7%	1.4%
en-US	mms	1500	1012	0.1784	[0.1682, 0.1882]	0.0605	30.6%	6.0%
es-ES	elevenlabs	1499	574	0.0269	[0.0229, 0.0316]	0.0087	82.7%	0.6%
es-ES	whisper	1499	574	0.0452	[0.0397, 0.0512]	0.0154	73.0%	0.9%
es-ES	mms	1499	574	0.1248	[0.1156, 0.1346]	0.0369	47.3%	3.7%
es-MX	elevenlabs	1500	639	0.0351	[0.0299, 0.0405]	0.0138	79.5%	1.0%
es-MX	whisper	1500	639	0.0613	[0.0541, 0.0691]	0.0214	68.3%	1.3%
es-MX	mms	1500	639	0.1482	[0.1358, 0.1610]	0.0471	40.9%	5.1%
nl-NL	elevenlabs	1500	236	0.0159	[0.0129, 0.0194]	0.0048	88.6%	0.0%
nl-NL	whisper	1500	236	0.0385	[0.0331, 0.0453]	0.0108	76.0%	0.5%
nl-NL	mms	1500	236	0.078	[0.0682, 0.0896]	0.0239	58.1%	1.3%
pt-BR	elevenlabs	1500	229	0.0531	[0.0441, 0.0631]	0.0169	78.0%	2.3%
pt-BR	whisper	1500	229	0.0737	[0.0608, 0.0884]	0.0234	71.5%	3.5%
pt-BR	mms	1500	229	0.2242	[0.1984, 0.2529]	0.0618	36.1%	12.9%
ky	elevenlabs	694	89	0.1829	[0.1486, 0.2268]	0.0647	47.7%	10.4%
ky	mms	694	89	0.2425	[0.2086, 0.2830]	0.058	25.6%	10.7%
ky	gigaam	694	89	0.0998	[0.0773, 0.1311]	0.0293	66.4%	5.5%
ru	whisper	1500	440	0.0543	[0.0473, 0.0628]	0.0148	71.8%	1.1%
ru	mms	1500	440	0.2259	[0.2107, 0.2429]	0.0467	24.8%	11.5%
ru	gigaam	1500	440	0.0481	[0.0419, 0.0559]	0.0116	72.2%	0.6%
kk	whisper	400	64	0.5097	[0.4667, 0.5538]	0.1253	6.8%	46.2%
kk	mms	400	64	0.2903	[0.2530, 0.3257]	0.0577	28.2%	16.8%
kk	gigaam	400	64	0.1179	[0.0930, 0.1417]	0.0263	57.8%	3.2%

Whisper has an internal language model; MMS uses greedy CTC without a language model. GigaAM provides the primary reading for Kyrgyz and Kazakh. Differences between recognisers are properties of the measurement instruments, not a cross-language ranking. Optional Scribe results exist only where measured; the hosted recogniser is not fully deterministic.

Recogniser agreement

lang	A	B	WER A	WER B	delta (A−B)	95% CI	identical transcripts	n shared
en-US	elevenlabs	whisper	0.0515	0.0774	-0.0259	[-0.0309, -0.0209]	68.2%	1500
en-US	elevenlabs	mms	0.0515	0.1784	-0.1269	[-0.1357, -0.1183]	32.1%	1500
en-US	whisper	mms	0.0774	0.1784	-0.1011	[-0.1086, -0.0934]	32.0%	1500
es-ES	elevenlabs	whisper	0.0269	0.0452	-0.0183	[-0.0227, -0.0143]	78.8%	1499
es-ES	elevenlabs	mms	0.0269	0.1248	-0.0979	[-0.1065, -0.0895]	49.0%	1499
es-ES	whisper	mms	0.0452	0.1248	-0.0796	[-0.0871, -0.0717]	50.0%	1499
es-MX	elevenlabs	whisper	0.0351	0.0613	-0.0262	[-0.0321, -0.0207]	73.9%	1500
es-MX	elevenlabs	mms	0.0351	0.1482	-0.1131	[-0.1240, -0.1028]	42.0%	1500
es-MX	whisper	mms	0.0613	0.1482	-0.0869	[-0.0952, -0.0788]	43.0%	1500
nl-NL	elevenlabs	whisper	0.0159	0.0385	-0.0226	[-0.0277, -0.0186]	78.9%	1500
nl-NL	elevenlabs	mms	0.0159	0.078	-0.0621	[-0.0730, -0.0540]	58.7%	1500
nl-NL	whisper	mms	0.0385	0.078	-0.0395	[-0.0478, -0.0328]	57.3%	1500
pt-BR	elevenlabs	whisper	0.0531	0.0737	-0.0207	[-0.0307, -0.0120]	73.7%	1500
pt-BR	elevenlabs	mms	0.0531	0.2242	-0.1712	[-0.1953, -0.1497]	36.2%	1500
pt-BR	whisper	mms	0.0737	0.2242	-0.1505	[-0.1682, -0.1333]	35.4%	1500
ky	elevenlabs	mms	0.1829	0.2425	-0.0596	[-0.0838, -0.0361]	20.3%	694
ky	elevenlabs	gigaam	0.1829	0.0998	0.0831	[0.0629, 0.1046]	43.1%	694
ky	mms	gigaam	0.2425	0.0998	0.1427	[0.1273, 0.1593]	24.8%	694
ru	whisper	mms	0.0543	0.2259	-0.1716	[-0.1849, -0.1599]	25.1%	1500
ru	whisper	gigaam	0.0543	0.0481	0.0062	[0.0014, 0.0106]	71.0%	1500
ru	mms	gigaam	0.2259	0.0481	0.1778	[0.1660, 0.1910]	25.4%	1500
kk	whisper	mms	0.5097	0.2903	0.2193	[0.1800, 0.2590]	6.5%	400
kk	whisper	gigaam	0.5097	0.1179	0.3918	[0.3531, 0.4320]	6.8%	400
kk	mms	gigaam	0.2903	0.1179	0.1725	[0.1473, 0.1966]	30.5%	400

Paired intervals resample the shared observations. Read the measured interval for each comparison; no conclusion is inferred from the sign of a point estimate alone.

Human anchors — speaker similarity

The anchor compares the prompt with a separate recording of the same speaker. The impostor floor compares distinct speakers. Mean floor, p95 and anchor - floor mean describe different aspects of speaker confusion.

lang	encoder	anchor	95% CI	floor mean	floor p95	usable range	n rows	speakers	distinct values
en-US	wavlm_sv	0.9317	[0.9283, 0.9347]	0.6119	0.8707	0.3198	909	631	629
en-US	wavlm_ft	0.652	[0.6424, 0.6601]	0.0725	0.2483	0.5795	909	631	630
en-US	ecapa	0.6141	[0.6042, 0.6227]	0.0822	0.2447	0.5319	909	631	631
es-ES	wavlm_sv	0.9435	[0.9392, 0.9469]	0.7286	0.9245	0.2149	1222	590	587
es-ES	wavlm_ft	0.6888	[0.6791, 0.6980]	0.1302	0.3183	0.5586	1222	590	590
es-ES	ecapa	0.6505	[0.6405, 0.6599]	0.132	0.3206	0.5185	1222	590	589
es-MX	wavlm_sv	0.9456	[0.9432, 0.9480]	0.7315	0.932	0.2141	1201	771	768
es-MX	wavlm_ft	0.6882	[0.6811, 0.6956]	0.1685	0.3576	0.5197	1201	771	769
es-MX	ecapa	0.635	[0.6268, 0.6438]	0.122	0.299	0.5131	1201	771	769
nl-NL	wavlm_sv	0.9285	[0.9235, 0.9331]	0.7446	0.9215	0.184	1319	412	412
nl-NL	wavlm_ft	0.6708	[0.6619, 0.6797]	0.174	0.352	0.4969	1319	412	412
nl-NL	ecapa	0.6232	[0.6134, 0.6336]	0.1866	0.3666	0.4366	1319	412	412
pt-BR	wavlm_sv	0.9339	[0.9265, 0.9403]	0.765	0.9248	0.1689	983	161	161
pt-BR	wavlm_ft	0.6346	[0.6166, 0.6502]	0.1864	0.3754	0.4482	983	161	161
pt-BR	ecapa	0.5763	[0.5578, 0.5930]	0.1257	0.2993	0.4506	983	161	161
ky	wavlm_sv	0.9267	[0.9122, 0.9381]	0.7355	0.9385	0.1912	599	155	154
ky	wavlm_ft	0.5795	[0.5584, 0.6002]	0.2207	0.422	0.3587	599	155	155
ky	ecapa	0.5641	[0.5413, 0.5855]	0.2111	0.4156	0.353	599	155	155
ru	wavlm_sv	0.9457	[0.9435, 0.9478]	0.7476	0.9394	0.1982	1383	1195	1185
ru	wavlm_ft	0.6684	[0.6626, 0.6741]	0.2692	0.4577	0.3992	1383	1195	1194
ru	ecapa	0.6348	[0.6283, 0.6411]	0.2187	0.4271	0.4161	1383	1195	1192
kk	wavlm_sv	0.9333	[0.9271, 0.9390]	0.7379	0.9389	0.1953	325	98	98
kk	wavlm_ft	0.5796	[0.5587, 0.6002]	0.2071	0.408	0.3724	325	98	98
kk	ecapa	0.5628	[0.5416, 0.5837]	0.2053	0.4039	0.3575	325	98	98

Use the matching language, encoder and reference domain in sim_norm = (sim_observed - sim_floor) / (sim_anchor - sim_floor). One reference pair is shared by a speaker's examples, so repeated row values are not independent measurements. Confidence intervals resample speakers.

encoder	model	purpose
`wavlm_sv`	`microsoft/wavlm-base-plus-sv`	continuity with historical results
`wavlm_ft`	`wavlm_large_finetune.pth`	seed-tts-eval comparison
`ecapa`	`speechbrain/spkrec-ecapa-voxceleb`	independent encoder family

WavLM-SV also supplies the candidate speaker-consistency filter; its scores therefore reflect selection in the same embedding space. ECAPA offers a separate reading. A model can exceed the human anchor because it is conditioned on the prompt recording; the anchor is a reference point rather than a ceiling.

SIM against prompt duration

lang	<3.7	3.7-4.5	>=4.5	monotonic
en-US	0.9228	0.9284	0.9367	True
es-ES	0.9321	0.9476	0.9456	False
es-MX	0.9377	0.9455	0.9492	True
nl-NL	0.925	0.9309	0.9367	True
pt-BR	0.9262	0.9491	0.9399	False
ky	0.9335	0.9294	0.8864	False
ru	0.9363	0.9431	0.9522	True
kk	0.9305	0.9406	0.9461	True

SIM against second-reference duration

lang	3.5-4.5	<3.5	>=4.5
en-US	0.948	0.9268	0.9432
es-ES	0.9528	0.9413	0.9572
es-MX	0.9519	0.9432	0.9574
nl-NL	0.9431	0.9279	0.8895
pt-BR	0.9451	0.9327	0.9376
ky	0.9174	0.9267	0.9508
ru	0.9547	0.9435	0.9581
kk	0.944	0.9318	0.9359

All encoders' breakdowns, cell counts and intervals are available in reports/csv/sim_breakdown.csv. Small speaker counts limit interpretation of apparent trends.

Reference domains and quality

Restored references: ky, ru, kk. Original Common Voice references: en-US, es-ES, es-MX, nl-NL, pt-BR. Ground-truth audio retains each subset's selected human recordings. The current delivered reference durations are:

subset	references	native prompt rates (Hz)	prompt duration min (s)	prompt duration max (s)
en-US	Common Voice	32000, 44100, 48000	2.016	7.296
es-ES	Common Voice	32000, 44100, 48000	2.016	6.624
es-MX	Common Voice	32000, 44100, 48000	2.016	7.008
nl-NL	Common Voice	32000, 44100, 48000	2.016	6.976
pt-BR	Common Voice	32000, 48000	2.016	7.168
ky	Sidon restored	48000	1.84	6.008
ru	Sidon restored	48000	0.576	7.168
kk	Sidon restored	48000	1.968	5.312

Both new-language reference streams are restored sequentially, one recording per Sidon call, in a separate environment. Native prompt outputs are retained in prompt_audio_orig; canonical reference columns are mono 16 kHz, VAD-trimmed and peak-normalised to −1 dBFS. New-language preparation uses soxr HQ and Silero VAD 6.2.3. The unchanged historical audio keeps its original build path.

lang	audio	n	DNSMOS OVRL	OVRL p05	narrowband %	low speech %	would fail prompt QC %
en-US	prompt	1500	2.761	1.971	0.27	0.27	0.53
en-US	gt	1500	2.766	1.968	1.47	0.93	2.4
en-US	sim_ref	909	2.752	2.028	0.22	0	0.22
es-ES	prompt	1500	2.766	2.011	1.33	0.2	1.53
es-ES	gt	1499	2.8	2.041	2.33	0.4	2.74
es-ES	sim_ref	1222	2.759	1.991	1.64	0.16	1.8
es-MX	prompt	1500	2.68	1.887	0.53	0	0.53
es-MX	gt	1500	2.662	1.834	2.73	1.8	4.27
es-MX	sim_ref	1201	2.672	1.861	1.25	0	1.25
nl-NL	prompt	1500	2.808	2.152	0.47	0	0.47
nl-NL	gt	1500	2.871	2.218	1.13	0.13	1.27
nl-NL	sim_ref	1319	2.804	2.082	1.59	0	1.59
pt-BR	prompt	1500	2.726	2.004	0.4	0	0.4
pt-BR	gt	1500	2.783	2.005	2.47	1.07	3.53
pt-BR	sim_ref	983	2.697	2.049	0.61	0.61	1.22
ky	prompt	700	3.147	2.691	1	0	1
ky	gt	694	2.749	1.924	1.87	0.58	2.31
ky	sim_ref	599	3.142	2.718	0	0.67	0.67
ru	prompt	1500	3.176	2.692	0.07	0	0.07
ru	gt	1500	2.823	2.12	1.87	0.67	2.4
ru	sim_ref	1383	3.145	2.667	0	0.07	0.07
kk	prompt	400	3.117	2.541	0	0	0
kk	gt	400	2.814	2.092	0.25	1.75	2
kk	sim_ref	325	3.088	2.524	0	0	0

DNSMOS measures signal/noise quality and is a reporting metric, not a selection filter or a naturalness score for synthesis. For restored subsets, prompt and second-reference QC describes delivered audio. Historical untrimmed prompt-QC values retain their existing meaning in the other five subsets. qc_bandwidth_hz is a deprecated alias of qc_hf_energy_db, measured in dB. Restored subsets have null qc_spk_centroid_dist, because their original candidate pools were not restored. Before/after SIM, quality and identity diagnostics: ky, ru, kk. Improved audio quality does not by itself establish preserved identity or better speaker separation.

Selection and preprocessing

S1 uses train/dev/test from Common Voice 17.0, deduplicated by clip path. Clips require at least two positive votes and no negative votes. Targets have 3 or more words, no digits, less than 5% letters outside the language alphabet and a source duration of 2–12 s. Prompts use a 3–8 s source window. eSpeak language-fallback markers cause rejection of affected target texts.

S2 uses the existing speech, clipping, bandwidth, duration, recording-continuity and speaker-consistency filters. Passing held-out clips are preferred; each speaker gets one fixed reference. S3 greedily maximises diphone coverage under length quotas, preserves available questions/exclamations, and assigns prompt slots under speaker caps and feasible female quotas. S4 assembles canonical audio and the separate same-speaker reference. Sidon follows selection and does not reselect examples, speakers or texts.

For the new languages, normalized prompt and target text must differ. Deterministic prompt-slot swaps repaired two Kazakh collisions while preserving target texts, coverage and the prompt-speaker/gender multiset. This verifies that the conditioning transcript is not reused as the target; it does not establish that the speaker never read that sentence elsewhere.

subset	examples	speaker cap	female target	length shares 3–5 / 6–9 / 10–12 / 13–16 / 17+	primary ASR	second ASR
ru	1500	2	20%	0.20 / 0.35 / 0.30 / 0.15 / 0.00	whisper	mms
kk	400	4	best effort	0.40 / 0.53 / 0.06 / 0.01 / 0.00	gigaam	mms

Stored audio is not VAD-trimmed again for evaluation. Synthesis uses mono, 16 kHz soxr HQ resampling, Silero VAD with 50 ms edge padding and −1 dBFS peaks. Whisper uses greedy short-form decoding without a repetition penalty; MMS uses the explicit language adapter and greedy CTC without a language model. GigaAM uses the pinned ctc revision. Exact run configurations, model revisions and package versions are supplied with each artifact.

Text normalisation applies NFC, lowercase, per-language digit expansion, punctuation removal and whitespace collapse on both reference and hypothesis. Apostrophes and diacritics are retained. num2words uses ru for Russian and kz for Kazakh. Primary per-row WER is not the corpus headline: sum shipped edit counts and divide by the summed reference lengths.

Release history

v3.1: new Russian and Kazakh subsets with restored references and measured human anchors; the six v3.0 subsets are preserved.

v3.0: Kyrgyz reference restoration changed its conditioning audio and dependent SIM/QC values while preserving its 700 examples, target text and ground-truth recordings. The other five subsets retained v2 data. The historical Kyrgyz evidence reports both audio-quality gains and speaker-separation losses; those findings apply to that fixed Kyrgyz selection.

v2.0: preserved v1 rows and audio, corrected macro-to-corpus aggregation, removed Whisper's repetition penalty and unnecessary long-form chunking, added MMS and independent speaker anchors, impostor floors, speaker-cluster intervals and auditable metric artifacts. The measured attribution table below concerns the six original subsets and their historical environment:

lang	ASR	published v1 (macro)	legacy (macro)	n legacy	nopen (macro)	n nopen	v2 (macro)	n v2	v2 (corpus)	v2 corpus 95% CI	drift (env)	D-02 penalty	agree legacy/nopen	agree nopen/v2	D-01 aggregation
en-US	whisper	0.0806	0.0805	1500	0.0807	1500	0.0807	1500	0.0774	[0.0703, 0.0842]	-0.0001	0.0002	0.9893	1	-0.0033
es-ES	whisper	0.0485	0.0485	1499	0.0484	1499	0.0484	1499	0.0452	[0.0397, 0.0512]	0	-0.0001	0.9907	1	-0.0032
es-MX	whisper	0.0654	0.0652	1500	0.0648	1500	0.0648	1500	0.0613	[0.0541, 0.0691]	-0.0002	-0.0004	0.9867	1	-0.0035
nl-NL	whisper	0.0407	0.0407	1500	0.0408	1500	0.0408	1500	0.0385	[0.0331, 0.0453]	0	0.0001	0.992	1	-0.0023
pt-BR	whisper	0.0823	0.0816	1500	0.0817	1500	0.0817	1500	0.0737	[0.0608, 0.0884]	-0.0007	0.0001	0.9727	1	-0.008
ky	gigaam	0.1104	0.1104	694	0.1104	694	0.1104	694	0.0998	[0.0773, 0.1311]	0	0	1	1	-0.0106

Data fields and reports

utt is the stable join key within a subset. has_gt and has_sim_ref describe audio availability. speaker_id identifies the prompt speaker; gt_speaker_id identifies the human reader. anchor_asr selects the primary recogniser. anchor_wer/anchor_cer and S/D/I/reference counts describe that recogniser's human read; optional MMS/Scribe fields describe their own reads. anchor_sim_* describes the fixed same-speaker reference pair under each encoder. phones is the documented voice's phone inventory, with native eSpeak units for Russian.

artifact	contents
`reports/coverage_*.md`	composition and candidate-pool coverage
`reports/csv/*.csv`	measured aggregate tables
`reports/per_utterance/*.parquet`	row metrics and reference lengths for reaggregation
`reports/per_utterance/*.json`	exact metric run configurations
`reports/extension/`	parent revision, new-language selection and validation evidence, source snapshot
`reports/restoration/`	preserved Kyrgyz restoration evidence
`reports/restoration/{ru,kk}/`	new reference hashes, environments, before/after metrics and embedding diagnostics

Limitations

No naturalness or expressive/prosodic evaluation; DNSMOS is reference-audio QC.
Short targets, at most 34 words in this release, with no digit or hard-text test set.
Crowd transcripts and readings may contain errors; ASR performance does not isolate them.
Anchors are comparable within the same language and metric configuration.
Speaker metadata is self-declared and often missing; Kazakh's source speaker pool is small.
Prompt reuse and uncapped GT-speaker concentration reduce independent sample size.
Common Voice may appear in recogniser or TTS training data; contamination is not quantified.
Restoration can change voice identity and impose shared artifacts; inspect the paired diagnostics.
Historical preprocessing is retained for existing audio; new languages use the documented soxr path.
Hosted optional ASR results may vary on repeated requests.

Reproduction, licence and citation

Download with msbench-fetch --revision v3.1. Rebuild the added languages with LANGUAGES_ru_kk.md, the source snapshot and recorded environments. The existing subsets do not need S1–S4 reruns. Metric configs allow reaggregation without model inference.

Dataset: CC0-1.0, matching Common Voice 17.0. Code: Apache-2.0. Model licences remain separate; MMS is non-commercial. Voices are volunteered human biometric data; speaker metadata is self-declared and identifiers are truncated hashes. The measurements assess voice cloning and do not authorise impersonation.

@misc{multilingual_speech_benchmark_2026,
  title = {Multilingual Speech Benchmark for Zero-Shot TTS},
  author = {nineninesix},
  year = {2026},
  note = {Version 3.1},
  url = {https://huggingface.co/datasets/nineninesix/multilingual-tts-benchmark}
}

Measurement details: PROTOCOL.md. Sources: Common Voice 17.0, Seed-TTS, Whisper, MMS, GigaAM-Multilingual, Sidon, WavLM, ECAPA-TDNN, DNSMOS P.835.

Downloads last month: 1,807

Papers for nineninesix/multilingual-tts-benchmark

WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing

Paper • 2110.13900 • Published Oct 26, 2021 • 3

DNSMOS P.835: A Non-Intrusive Perceptual Objective Speech Quality Metric to Evaluate Noise Suppressors

Paper • 2110.01763 • Published Oct 5, 2021