More than ten million hours of speech with precisely aligned transcripts, covering Chinese, English, low-resource languages and dialects, for both recognition and synthesis.
What makes it worth training on.
10M+ hours, including languages where public data is thin.
Undergraduate STEM: 40M+ today, 100M target →Precise text–audio alignment, ready for ASR and TTS training.
100% Markdown/LaTeX · schema-checked JSONL →Choose languages, dialects, accents and audio specifications.
Scope → pilot → acceptance → scale →Samples on request.
Tell us the languages, dialects and task (ASR or TTS), and we will send a matched audio sample pack.
Request a sample →How it is sliced.
Languages
Chinese and English; low-resource languages across Europe, Japan & Korea and Southeast Asia; regional dialects.
ASR
Multilingual speech → text, across languages, dialects and accents.
TTS
Text → multilingual speech, across languages, dialects and accents.
Alignment
High-accuracy text–audio correspondence for cross-language and cross-accent adaptation.
Record format.
| Field | Type | Description |
|---|---|---|
| audio | file | Audio segment (format and sample rate to specification). |
| transcript | string | Aligned transcript. |
| language / dialect | string | Language and dialect tags. |
| segments | array | Optional timestamped alignment. |
| speaker | object | Optional speaker attributes. |
Built for.
- ASR modelsMultilingual and dialect recognition.
- TTS modelsMultilingual and dialect synthesis.
- Low-resource languagesCoverage where public data is thin.
Other linesfrom the same engine.
Request a pilot slice.
Tell us the multilingual speech slice you need. We will send a pilot sample cut to your specification for your own evaluation.