Add Gnani (STT and TTS) community integrations - #991
Conversation
|
|
||
| ## Overview | ||
|
|
||
| `GnaniHttpSTTService` and `GnaniSTTService` transcribe speech using |
There was a problem hiding this comment.
Would it be simpler to just add support for the streaming services? Or, are there use cases where the HTTP API is better suited for voice agents?
Same question for the TTS services.
There was a problem hiding this comment.
Hi @markbackman , For STT, we already have a streaming service in place, and for TTS we have an SSE-based output streaming service, so the streaming path is fully supported on both sides.
the REST-based approach is more economical. Our latency on the REST path is also good, making it well-suited to voice agents.
There was a problem hiding this comment.
Makes sense. Include services that are best suited for realtime.
There was a problem hiding this comment.
@markbackman Thanks for the review! We'd prefer to keep both the realtime and HTTP paths if possible. A significant part of our user base is in India, where cost is a major consideration, so the HTTP option is valuable for us. Our model has also been benchmarked with very low latency, so the tradeoff there is minimal.
Could you help move this PR forward when you get a chance? Happy to make any changes you'd like before merging.
…uage parameter and clarify features
| "source": "/server/services/stt/gladia", | ||
| "destination": "/api-reference/server/services/stt/gladia" | ||
| }, | ||
| { |
There was a problem hiding this comment.
No need for a redirect since these were just added.
There was a problem hiding this comment.
sure , will remove those
| "source": "/server/services/tts/fish", | ||
| "destination": "/api-reference/server/services/tts/fish" | ||
| }, | ||
| { |
markbackman
left a comment
markbackman
left a comment
There was a problem hiding this comment.
Before moving forward, I think it's worth considering which APIs to expose to Pipecat. Do you want to maintain HTTP and Websocket endpoints? Pipecat is for realtime / voice use cases, so streaming services are almost always preferred.
markbackman
left a comment
markbackman
left a comment
There was a problem hiding this comment.
LGTM! Thanks for adding your services 🙌
2 participants
Gnani
Adds Gnani to the Community Integrations page under both Speech-to-Text and Text-to-Speech.
Gnani Vachana is a production speech AI platform for Indian languages, providing high-accuracy STT and low-latency TTS across REST, SSE, and WebSocket. The integration includes
GnaniHttpSTTService,GnaniSTTService,GnaniHttpTTSService,GnaniSSETTSService, andGnaniTTSService. WebSocket STT supports 10+ Indian languages at 8/16/44.1/48 kHz, while TTS offers four voices (Pranav, Kaveri, Shubhra, and Deepak) with interruptible WebSocket streaming.Tested with the latest Pipecat main branch (v1.5.0+).
Repository: https://github.com/Gnani-AI-Mintlify/pipecat-gnani
PyPI: https://pypi.org/project/pipecat-gnani/
Core SDK: https://pypi.org/project/gnani-vachana/
Demo Video: https://go.screenpal.com/watch/cOi2nynUax5
Maintainers: (Gnani.ai)