skit-ai / speech-recognition

SDKs and docs for Skit's speech to text service

Speech-to-Text API

Converts audio to text

We support these ten indian languages (language codes).

• Hindi
• English
• Marathi
• Kannada
• Malayalam
• Bengali
• Gujarati
• Punjabi
• Telugu
• Tamil

Authentication

To get access to our APIs reach out to us at [email protected] We do not provide public access token for the APIs anymore.

Ways to use the Service

• Transcribing short audios [audios upto 1 min]
• Transcribing long audios [more than 1 min]
• Transcribing audio from streaming input

We recommend that you call this service using Vernacular provided client libraries. If your application needs to call this service using your own libraries, you should use the HTTP Endpoints.

Supported SDKs: Python

REST Reference

ServiceHost: https://asr.vernacular.ai

Speech Recognition

Name	Description
recognize	Performs synchronous speech recognition: receive results after all audio has been sent and processed.
longrunningrecognize	Performs asynchronous speech recognition. Generally used for long audios

RPC Reference

Speech Recognition

Methods	Description
Recognize	Performs synchronous speech recognition: receive results after all audio has been sent and processed.
LongRunningRecognize	Performs asynchronous speech recognition: receive results via the longrunning.Operations interface.
StreamingRecognize	Performs streaming speech recognition: receive results while sending audio. Supports both unidirectional and bidirectional streaming.

speech-recognition's Issues

Making more models accessible

There are many experimental kaldi-serve models (like phoneme decoders) that are now being used widely in the team. What kind of effort would be needed make these models accessible through this API ?