tiny-aya-translate/cv-tr-eval
Common Voice Turkish Eval 4,825 Turkish test clips (~45 MB, 16 kHz) in the Mozilla Common Voice schema: transcription, duration, up_votes / down_votes, and the age / gender / accent speaker attributes. Schema in the YAML header above. An evaluation-only Turkish counterpart to lahaja-eval; never trained on. Used to sanity-check Turkish ASR quality on real human speech, which matters here because the v0.3 training corpus is entirely synthetic TTS and the released model is… See the full description on the dataset page: https://huggingface.co/datasets/tiny-aya-translate/cv-tr-eval.
Common Voice Turkish Eval
4,825 Turkish test clips (~45 MB, 16 kHz) in the Mozilla Common Voice schema: transcription, duration, up_votes / down_votes, and the age / gender / accent speaker attributes. Schema in the YAML header above.
An evaluation-only Turkish counterpart to `lahaja-eval`; never trained on. Used to sanity-check Turkish ASR quality on real human speech, which matters here because the v0.3 training corpus is entirely synthetic TTS and the released model is measurably distribution-bound.
Derived from Mozilla Common Voice; the upstream release terms apply.
Code
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).