Dataset Viewer
The dataset viewer is not available for this split.
Cannot load the dataset split (in streaming mode) to extract the first rows.
Error code: StreamingRowsError
Exception: CastError
Message: Couldn't cast
latency_s: double
lines: list<item: struct<at: double, confirmed: null, id: int64, is_final: bool, original: string, pipeline (... 101 chars omitted)
child 0, item: struct<at: double, confirmed: null, id: int64, is_final: bool, original: string, pipeline_latency_ms (... 89 chars omitted)
child 0, at: double
child 1, confirmed: null
child 2, id: int64
child 3, is_final: bool
child 4, original: string
child 5, pipeline_latency_ms: int64
child 6, provider: string
child 7, source_lang: string
child 8, translated: string
child 9, translated_2: string
wav: string
license: string
seconds: double
id: string
source: string
lang: string
part: string
segments: list<item: struct<start: double, end: double>>
child 0, item: struct<start: double, end: double>
child 0, start: double
child 1, end: double
to
{'id': Value('string'), 'part': Value('string'), 'lang': Value('string'), 'source': Value('string'), 'license': Value('string'), 'seconds': Value('float64'), 'segments': List({'start': Value('float64'), 'end': Value('float64')})}
because column names don't match
Traceback: Traceback (most recent call last):
File "/src/services/worker/src/worker/utils.py", line 149, in get_rows_or_raise
return get_rows(
dataset=dataset,
...<4 lines>...
column_names=column_names,
)
File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
return func(*args, **kwargs)
File "/src/services/worker/src/worker/utils.py", line 129, in get_rows
rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
File "/src/services/worker/src/worker/utils.py", line 544, in safe_iter
yield from ds.decode(False) if ds.features else ds
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 3052, in __iter__
for key, example in ex_iterable:
^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
for key, pa_table in self._iter_arrow():
~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, in _iter_arrow
for key, pa_table in self.ex_iterable._iter_arrow():
~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 558, in _iter_arrow
for key, pa_table in iterator:
^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, in _iter_arrow
for key, pa_table in self.generate_tables_fn(**gen_kwags):
~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 361, in _generate_tables
self._cast_table(pa_table, json_field_paths=json_field_paths),
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
pa_table = table_cast(pa_table, self.info.features.arrow_schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2381, in table_cast
return cast_table_to_schema(table, schema)
File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
raise CastError(
...<3 lines>...
)
datasets.table.CastError: Couldn't cast
latency_s: double
lines: list<item: struct<at: double, confirmed: null, id: int64, is_final: bool, original: string, pipeline (... 101 chars omitted)
child 0, item: struct<at: double, confirmed: null, id: int64, is_final: bool, original: string, pipeline_latency_ms (... 89 chars omitted)
child 0, at: double
child 1, confirmed: null
child 2, id: int64
child 3, is_final: bool
child 4, original: string
child 5, pipeline_latency_ms: int64
child 6, provider: string
child 7, source_lang: string
child 8, translated: string
child 9, translated_2: string
wav: string
license: string
seconds: double
id: string
source: string
lang: string
part: string
segments: list<item: struct<start: double, end: double>>
child 0, item: struct<start: double, end: double>
child 0, start: double
child 1, end: double
to
{'id': Value('string'), 'part': Value('string'), 'lang': Value('string'), 'source': Value('string'), 'license': Value('string'), 'seconds': Value('float64'), 'segments': List({'start': Value('float64'), 'end': Value('float64')})}
because column names don't matchNeed help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
ucast.me Local-mode benchmark
All the data of the benchmark the ucast.me desktop app's Local mode is measured on: speech in, live
subtitles (transcript + translations) out, through the app's own real-time pipeline. Kept in one place
so every run can be compared with the next one. Built and run with the scripts in the app repository,
training/benchmark/ (prep_*.py, run.py, score.py, publish.py).
What is here: only what the sources' licences let anyone redistribute. The audio and references of
FLEURS, EdAcc, AMI, CIEMPIESS, Common Voice Spontaneous and the app's own Pocket TTS test talks are here;
mTEDx and CORAA (non-commercial, no derivatives), Europarl-ST (non-commercial) and what is made from them
(room, long) are here as recipes — source, licence, length and each segment's time, no audio or
text — that the prep_*.py scripts rebuild from the source. Scores and reports cover all of them; the subtitles the app wrote are here for the redistributable
items only.
Usage scenarios
Each folder of sets/ is one situation the app's subtitles run in (catalog: scenarios.json in the
app's repository, training/benchmark/):
| scenario | what it is | languages | translations to score |
|---|---|---|---|
presentation |
a talk with a good microphone: mTEDx (5 min of continuous audio of 2 talks per language) and Europarl-ST (whole parliament speeches, ~10 min per language) | mTEDx: es fr pt it ru de · Europarl-ST: en pl nl | mTEDx: es/fr/pt/it/ru → en; Europarl-ST: between en fr de it es pt pl ro nl |
room |
the same talks heard ~6 m away in a simulated lecture room (RT60 0.8 s, air-conditioning noise at 15 dB SNR) | en pt fr es | as presentation |
accent |
English said by non-native speakers (EdAcc conversations, 8 speakers with different L1s) | en | none |
technical |
a technical talk with its slide (names, products, acronyms; synthetic Pocket TTS voices) | pt fr | none — scores the terms' exact spelling |
bilingual |
a dialogue in two languages (FLEURS sentences alternated), run with auto-swap | pt-fr en-es | each sentence into the other language |
long |
30 minutes of whole mTEDx talks in a row | pt | en |
spontaneous |
unscripted speech: meetings (AMI), interviews/conversations (CORAA), radio (CIEMPIESS), free answers (Common Voice Spontaneous) | en pt es de | none (no human translation exists) — transcript only |
read |
FLEURS test: one read Wikipedia sentence per clip, 50 clips per language, played 4 s apart (regression only) | en fr pt es de it nl pl ru uk | every other app language (FLEURS is parallel) |
Layout
sets/<part>/<lang>/<item>.wav 16 kHz mono 16-bit
sets/<part>/<lang>/<item>.json reference transcript, human translations, source, licence,
and for long items the segments ("clips") with their times
recipes/<part>/<lang>/<item>.json the non-redistributable items: source, licence, seconds, segment times
runs/<run>/<part>/<lang>/out.jsonl every subtitle the app showed (drafts and finals, with times; redistributable items)
runs/<run>/<part>/<lang>/settings.json engine, models, commit, machine, subtitle languages
runs/<run>/<part>/<lang>/scores.json WER (eaten/invented), chrF per subtitle language, lag, latency
runs/<run>/scores.json all of them; reports/<run>.md the write-up
Sources and licences
| source | used for | licence |
|---|---|---|
| google/fleurs test | read | CC-BY-4.0 |
| Multilingual TEDx test (Salesky et al. 2021), original archives | presentation | CC-BY-NC-ND-4.0 |
| Europarl-ST v1.1 test (Iranzo-Sánchez et al. 2020), segment list from tj-solergibert/Europarl-ST | presentation | CC-BY-NC-4.0 |
| edinburghcstr/edacc test (Sanabria et al. 2023) | accent | CC-BY-SA-4.0 |
babelcast rust/desktop-egui/tests/fixtures/speech_terms_* (Pocket TTS, Kyutai) + the app's OCR of the slide |
technical | CC-BY-4.0 |
| edinburghcstr/ami ihm test | spontaneous (en) | CC-BY-4.0 |
| gabrielrstan/CORAA-v1.1 test, "Spontaneous Speech" segments | spontaneous (pt) | CC-BY-NC-ND-4.0 |
| ciempiess/ciempiess_test | spontaneous (es) | CC-BY-SA-4.0 |
| q1805/common_voice_spontaneous_speech_4_0_de | spontaneous (de) | CC-BY-4.0 (Common Voice data CC0) |
Each file here keeps its source's licence (the CIEMPIESS excerpts are CC-BY-SA-4.0 like the source; the scores, reports and the app's subtitles are ours, CC-BY-4.0). mTEDx and Europarl-ST items are the continuous recording (Europarl-ST: only its transcribed stretches, gaps kept up to 2 s); AMI's headset channels are mixed back at their times; CORAA, CIEMPIESS and Common Voice clips are joined in order. Nothing was transcribed or translated by us.
- Downloads last month
- 111