Datasets:

tiny-aya-translate
/

tr-hi-parallel-speech-v2

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
pair_id: large_string
voice: large_string
tr_text: large_string
hi_text: large_string
en_text: large_string
tr_audio: large_string
hi_audio: large_string
tr_duration_s: double
hi_duration_s: double
source: large_string
-- schema metadata --
pandas: '{"index_columns": [], "column_indexes": [], "columns": [{"name":' + 1225
to
{'audio_id': Value('string'), 'pair_id': Value('string'), 'src_lang': Value('string'), 'tgt_lang': Value('string'), 'src_text': Value('string'), 'tgt_text': Value('string'), 'en_text': Value('string'), 'audio': Audio(sampling_rate=None, decode=True, num_channels=None, stream_index=None), 'tts_model': Value('string'), 'tts_voice': Value('string'), 'duration_s': Value('float64'), 'wer': Value('float64'), 'cer': Value('float64'), 'source': Value('string')}
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 489, 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 2818, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2355, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2380, 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 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, 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/parquet/parquet.py", line 220, in _generate_tables
                  yield Key(file_idx, batch_idx), self._cast_table(pa_table)
                                                  ~~~~~~~~~~~~~~~~^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/parquet/parquet.py", line 156, 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 2369, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              pair_id: large_string
              voice: large_string
              tr_text: large_string
              hi_text: large_string
              en_text: large_string
              tr_audio: large_string
              hi_audio: large_string
              tr_duration_s: double
              hi_duration_s: double
              source: large_string
              -- schema metadata --
              pandas: '{"index_columns": [], "column_indexes": [], "columns": [{"name":' + 1225
              to
              {'audio_id': Value('string'), 'pair_id': Value('string'), 'src_lang': Value('string'), 'tgt_lang': Value('string'), 'src_text': Value('string'), 'tgt_text': Value('string'), 'en_text': Value('string'), 'audio': Audio(sampling_rate=None, decode=True, num_channels=None, stream_index=None), 'tts_model': Value('string'), 'tts_voice': Value('string'), 'duration_s': Value('float64'), 'wer': Value('float64'), 'cer': Value('float64'), 'source': Value('string')}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

TR↔HI Parallel Speech (v2) — synthetic TTS corpus

The raw speech corpus behind TinyAya Stage 2: ~911 hours of synthetic Turkish⇄Hindi parallel speech, 53,506 rows, generated with OmniVoice across 14 voice designs.

This is the pre-encoding source. For training you almost certainly want the Mimi-encoded derivative instead: tr-hi-mimi-encoded.

Layout

path	contents
`data/train-*.parquet`	the loadable table (schema in the YAML header above)
`audio/*.wav`	~9,979 individual clips
`audio_tars/`	46 packed tarballs of the same audio
`manifests/`, `manifests_filtered/`	generation + QC manifests

Each row carries the text pair (src_text/tgt_text, plus en_text pivot), the TTS model and voice, duration, and round-trip ASR quality signals (wer, cer) — so you can apply your own quality threshold.

Quality control

Clips were validated by round-trip ASR (faster-whisper large-v3): transcribe, then score WER against the prompt text. The accepted band is WER ≤ 0.20 with duration 0.5–30 s; overall pass rate was 86%. wer/cer are retained per row rather than pre-filtered, so nothing is silently dropped for you.

⚠️ Synthetic speech. Every clip is TTS output, not a human recording. Models trained on it are distribution-bound — the v0.3 evaluation measured a sharp drop on real human speech (FLEURS).

Where this sits

The v0.3 speech-to-speech pipeline, end to end:

tr-hi-parallel-text          text triples (en pivot -> tr / hi)
        |  TTS
tr-hi-parallel-speech-v2     synthetic speech + QC signals
        |  Mimi encode
tr-hi-mimi-encoded           8-codebook tokens + word alignments
        |  Stage-2 training
tr-hi-s2st-v0.3              the released model


Model	`tr-hi-s2st-v0.3`
Text	`tr-hi-parallel-text`
Speech	`tr-hi-parallel-speech-v2` · `-v3`
Encoded	`tr-hi-mimi-encoded`
Eval sets	`fleurs-tr-hi-mimi-encoded` · `lahaja-eval` · `cv-tr-eval`

Code

repo	what it does
`sound-quality-check`	4-stage speech-dataset quality control
`model`	Stage-2 training, evaluation harness and TPU launch tooling

Project

TinyAya Stage 2 — Turkish⇄Hindi speech-to-speech translation with a text inner-monologue: a LoRA-adapted Cohere2 backbone driving a frozen Moshi depth decoder over Mimi codes.

The v0.3 run covered 76,250 steps / 2.07 epochs on a Cloud TPU v6e-16 (best val composite 2.8199 @ step 76,000). Read honestly: the text inner-monologue learns to translate (free-run chrF++ ~25.7 / 25.1), while intelligible audio synthesis remains the frontier (ASR-chrF++ 3.7 / 9.6 against a 92.1 / 86.6 ground-truth-audio ceiling) — bounded by the frozen depth decoder, not by translation understanding.

Results: v0.3 evaluation report
Training run: W&B xzcb60bl · emergence report
Blog: Adapting Moshi for Low-Resource Speech Translation

Compute for the v0.3 run was provided by Google's TPU Research Cloud (TRC).

Downloads last month: 436