translation_dataset

Synthetic, expressive, multilingual speech for cross-lingual dubbing research. Each example pairs a style-annotated text with generated audio that clones an English reference voice: the voice stays the same, the language changes.

• ~1.9M examples in the one_speaker config
• 17 languages
• ~900 distinct reference speakers

Samples

Each sample shows the generated audio followed by the English reference voice that conditioned it.

English (en)

Italian (it)

Russian (ru)

Arabic (ar)

Chinese (zh)

Japanese (ja)

Korean (ko)

Hindi (hi)

Configs

All nine subsets below are registered as loadable configurations with a train split.

Config	Content	Parquet files
one_speaker	Single-speaker synthetic speech with reference audio and text	1,913
multispeaker	Multispeaker synthetic speech with reference audio and text	537
ambient_v3	Ambient-background audio with transcription, prompt and profile	241
ambient_v4	Regenerated ambient backgrounds; generation is currently stopped	358
music_sfx_step2	Multispeaker scenes with separate voice, music, SFX, background and mixed audio	1,053
dynamic_sfx_step2	Multispeaker scenes with dynamic SFX and separate audio stems	1,052
music_sfx_step3	Available Step 2 music scenes with regenerated backgrounds	80
dynamic_sfx_step3	Available Step 2 dynamic-SFX scenes with regenerated backgrounds	54
wavcaps_bg	Recorded BBC + Freesound backgrounds in one tagged table, including crowd/babble	216

File counts reflect the repository snapshot inspected on 2026-09-26; they are not row counts. Some generation runs were stopped before exhausting their source data, so Step 3 and ambient_v4 should not be assumed to cover all rows in earlier versions. WavCaps publication is complete: 33,409 recordings, approximately 1,959 hours.

Load any subset

from datasets import load_dataset

ds = load_dataset(
    "mlinmg/translation_dataset", "multispeaker", split="train", streaming=True
)
# Replace "multispeaker" with any config name in the table above.

Multispeaker and background fields

multispeaker contains id (string), language, transcription, references (a list of reference audio bytes and text), and generated_audio (bytes/path). The original one_speaker IDs are integers; do not assume all configs share an ID type.

The four music_sfx_step2/3 and dynamic_sfx_step2/3 configs additionally contain duration_seconds, n_speakers, bg_prompt, bg_profile, and separate generated_audio, sfx_audio, music_audio, bg_audio, mixed_audio columns. Their references also retain speaker_id, source, revision, and sha256.

ambient_v3 and ambient_v4 contain id, language, transcription, bg_prompt, bg_profile, and bg_audio. They do not expose the separate multispeaker stems. wavcaps_bg has a distinct schema with source, tags, caption, and provenance; see its dedicated section below. Schemas are declared in the YAML dataset_info.

Structure (one_speaker)

field	type	description
id	int	utterance ID; the same id appears once per language, linking parallel translations
language	string	ISO 639-1 code
transcription	string	text with inline style directives
references	list<struct>	reference voices, {audio, text}
generated_audio	Audio	synthesized speech

Background configs add fields such as bg_prompt, bg_profile, bg_audio, and, for multi-speaker configs, separate sfx_audio, music_audio and mixed_audio stems.

Style directives

Transcriptions contain bracketed prosodic cues that are meant to be performed, not read:

[inhale] So, the detective... [soft voice] just looked at the file
[long pause] and then he [gasp] he said it. [angry] It w-w-wasn't a robbery.
[shouting] It was an ambush!!!

Stuttering, syllabic emphasis and repeated punctuation are also used as intensity cues. These cues are requests to the synthesizer and are not verified perceptual annotations.

Loading

from datasets import load_dataset

ds = load_dataset("mlinmg/translation_dataset", "one_speaker", split="train", streaming=True)
ex = next(iter(ds))

The dataset is several hundred GB; streaming is recommended.

Limitations

• Synthetic configs: their audio is synthetic and should not be treated as recorded human speech. wavcaps_bg contains recorded-source assets; see its section below.
• English-only reference voices, so target-language accents may be influenced by the source timbre.
• Intelligibility, speaker fidelity and background quality have not been exhaustively validated. Synthesis artifacts may be present.
• License to be determined.

Recorded WavCaps backgrounds (wavcaps_bg)

One combined train table containing BBC and Freesound background candidates from cvssp/WavCaps at revision 0930ec11ded28fa0eaa910fde2f6fc3538acbeac. These are recorded-source assets, distinct from the synthetic configs above. Publication is complete; wavcaps_bg/PUBLICATION_COMPLETE.json records the verified shards.

bg_audio contains the original encoded audio, with no resampling or normalization. id is prefixed by source; source is bbc or freesound. Full source metadata and original IDs are retained in source_metadata_json. duration is measured in seconds; sha256 hashes the original encoded audio; source_revision pins provenance.

Filter using the list-valued tags column: nature, urban, indoor_mechanical, crowd (including chatter/babble), ambience, and music_mentioned. Tags come from source captions/metadata, not an acoustic classifier. Crowd, babble and environmental music are allowed. Candidates are at least 15 seconds long. Decoded audio integrity and byte preservation are checked, but this does not certify perceptual quality or speech-free backgrounds: acoustic_background_validation is false for every row.

from datasets import load_dataset
ds = load_dataset("mlinmg/translation_dataset", "wavcaps_bg", split="train", streaming=True)
crowd = ds.filter(lambda row: "crowd" in row["tags"])
bbc = ds.filter(lambda row: row["source"] == "bbc")