[REKA SPEECH]

Meaning Made Intelligible 

Meaning Made Intelligible 

Meaning Made Intelligible 

Reka Speech is an on-device multilingual speech-to-text and speech-to-speech transcription and translation model.

Reka Speech is an on-device multilingual speech-to-text and speech-to-speech transcription and translation model.

Reka Speech is an on-device multilingual speech-to-text and speech-to-speech transcription and translation model.

Try it Now

Try it Now

Try it Now

Schedule a Demo

Schedule a Demo

Schedule a Demo

Joing our Discord

Joing our Discord

Joing our Discord

[SPECIFICATIONS]

Ultra compact and versatile, for embedding speech AI into the smallest devices

Ultra compact and versatile, for embedding speech AI into the smallest devices

Ultra compact and versatile, for embedding speech AI into the smallest devices

Runs on cellphones, cars, robots, and smart appliances.

~900 million parameters

English, Japanese, Korean, Chinese, French, Spanish, German, Italian, Portuguese, Arabic

Less than 1GB memory size

Also available via Reka API

~900 million parameters

English, Japanese, Korean, Chinese, French, Spanish, German, Italian, Portuguese, Arabic

Less than 1GB memory size

Also available via Reka API

~900 million parameters

English, Japanese, Korean, Chinese, French, Spanish, German, Italian, Portuguese, Arabic

Less than 1GB memory size

Also available via Reka API

This model can be customized further. Chat with our team ↗

This model can be customized further.

Chat with our team ↗

[PERFORMANCE]

State-of-the-art performance

Outperforms popular open source models such as Whisper-L v3 which is three times larger on transcription and translation quality.

Translation accuracy (Higher is better)

A horizontal bar chart comparing the performance of Reka Spark against Whisper L V3 across eight language pairs (Korean, Japanese, Chinese, French, Italian, Spanish, German, and Portuguese to English). Reka Spark consistently outperforms Whisper L V3 in every category, with scores ranging from 78.6 to 96.0.
A horizontal bar chart comparing the performance of Reka Spark against Whisper L V3 across eight language pairs (Korean, Japanese, Chinese, French, Italian, Spanish, German, and Portuguese to English). Reka Spark consistently outperforms Whisper L V3 in every category, with scores ranging from 78.6 to 96.0.
A horizontal bar chart comparing the performance of Reka Spark against Whisper L V3 across eight language pairs (Korean, Japanese, Chinese, French, Italian, Spanish, German, and Portuguese to English). Reka Spark consistently outperforms Whisper L V3 in every category, with scores ranging from 78.6 to 96.0.

Transcription error rate (Lower is better)

A horizontal bar chart comparing speech recognition error rates between Whisper L V3 and Reka Spark across nine languages (English, Korean, Japanese, Chinese, French, Italian, German, Spanish, and Portuguese). Reka Spark consistently shows lower error rates than Whisper L V3 in almost every category.
A horizontal bar chart comparing speech recognition error rates between Whisper L V3 and Reka Spark across nine languages (English, Korean, Japanese, Chinese, French, Italian, German, Spanish, and Portuguese). Reka Spark consistently shows lower error rates than Whisper L V3 in almost every category.
A horizontal bar chart comparing speech recognition error rates between Whisper L V3 and Reka Spark across nine languages (English, Korean, Japanese, Chinese, French, Italian, German, Spanish, and Portuguese). Reka Spark consistently shows lower error rates than Whisper L V3 in almost every category.
3D clusters of black and pink cubes with the Reka "[R]" logo, representing the modular components of multimodal AI models.

Reka Vision is an intelligence layer that transforms multimodal data into insights and actions

Contact Us

3D clusters of black and pink cubes with the Reka "[R]" logo, representing the modular components of multimodal AI models.

Reka Vision is an intelligence layer that transforms multimodal data into insights and actions

Contact Us

3D clusters of black and pink cubes with the Reka "[R]" logo, representing the modular components of multimodal AI models.

Reka Vision is an intelligence layer that transforms multimodal data into insights and actions

Contact Us