Navana Streaming ASR Instructions

How to Stream

Connection Instructions

• Endpoint: Websocket streaming speech API endpoint

  • wss://bodhi.navana.ai

• Sample Script:

  • streaming_client.js (for static audio files)
  • streaming_client_with_conversion.js (for static audio files)

Access Token

Store the authentication headers in env to access the streaming speech API endpoints:

$ export API_KEY=YOUR_API_KEY
$ export CUSTOMER_ID=YOUR_CUSTOMER_ID

Brief Description of Response Format

The received response format will be a JSON object.

{
  "call_id": "CALL_ID",
  "segment_id": "SEGMENT_ID",
  "eos": false,
  "type": "partial",
  "text": "TRANSCRIPT"
}

Keys Description

• Call_id:

  • Unique identifier associated with every streaming connection

• Segment_id:

  • Unique identifier associated with every speech segment during the entire active socket connection

• Text:

  • If type = "partial"
    • Partial transcript corresponding to every streaming audio chunk
    • Partial transcripts for every audio chunk (will be for a 100ms audio chunk if streaming audio packet size is 100ms)
  • If type = "complete"
    • Complete/final transcript generated for each speech segment
    • Generated once per segment_id i.e., when the speech segment end is reached

• eos:

  • If 'eos' is true, marks the end of the streaming connection

Install packages

npm install

Usage

Streaming Client with Audio File (Mono, 8kHz)

node streaming_client.js -f loan.wav

Options: -f: File name of the audio file to be streamed.

Streaming Client with On-the-Fly Audio Conversion

node streaming_client_with_conversion.js

Audio Stream Requirements

To ensure optimal compatibility and performance with our audio processing system, please adhere to the following audio stream requirements:

• Encoding/Bit Depth: 16Bit PCM with a 2 Byte depth, providing high-quality audio representation.

• Minimum Sample Rate: The audio must have a sample rate of at least 8000Hz.

• Fixed Streaming Rate: Audio packets should be streamed at (chunk_duration_ms) a fixed size (50 - 500 ms), ensuring consistent data flow. We recommend using 100 ms as shown in the example script.

• Channels: Audio must be single-channel (Mono) to ensure compatibility with our processing pipeline.

• Speakers: Initially, support is provided for a single speaker per channel. However, support for multiple speakers on a single channel is under development and will be announced soon.

Available ASR Models for Testing

• Hindi: hi-general-v2-8khz
• Kannada: kn-general-v2-8khz

For testing the code, modify the .js file with the model name you want to use.