- Home
- Training data
- Audio & Speech
Audio and speech training data.
Licensed recordings for speech recognition, text-to-speech, and voice agents, delivered with transcripts, speaker labels, and quality scores. Off the shelf where we have it, recorded to your spec where we don't.
What we can supply
- Conversational and multi-party speech, including meetings and calls
- Accented, dialect, and code-switched speech
- Expressive and emotional speech for text-to-speech
- Low-resource and endangered languages, validated by native speakers
- Far-field and noisy-environment recordings
- Read speech and long-form audio: podcasts, broadcasts, and audiobooks
- Domain speech, such as clinical conversations, fully de-identified
Labels and enrichment
- Verbatim transcripts
- Word-level timing
- Speaker diarization
- Language and dialect ID
- Emotion and intensity
- Intent and sentiment
- Phoneme alignment
- Audio quality scores
- Isolated vocal stems
Formats and delivery
WAV or FLAC audio with JSON or JSONL transcripts and metadata. Delivered to AWS S3, Google Cloud Storage, Azure, or your own storage, with a datasheet and provenance records.
Common uses
- Speech recognition
- Text-to-speech and voice cloning
- Voice agents
- Speaker diarization
- Emotion recognition
- Speech enhancement
- Audio language models
Questions
Which languages can you supply?
We license and record speech in high-, mid-, and low-resource languages, with native-speaker validation. Tell us your target languages and we confirm coverage before anything is quoted.
Can you record to our own script or conditions?
Yes. Custom collection covers scripted prompts, specific devices and acoustic conditions, and the speaker demographics you need.
Is voice cloning covered by the license?
Tell us the intended use up front. Voice cloning and synthetic-voice uses are scoped explicitly in the license so the consent behind the data matches how you use it.
Own data like this? License it through ModelCore