unnecessarily long pauses

[jpage-bridge]

in this code block: https://github.com/twilio-labs/call-gpt/blob/main/services/transcription-service.js#L46

// console.log(text, "is_final: ", transcription?.is_final, "speech_final: ", transcription.speech_final);
// if is_final that means that this chunk of the transcription is accurate and we need to add it to the finalResult 
if (transcriptionEvent.is_final === true && text.trim().length > 0) {
  this.finalResult += ` ${text}`;
  // if speech_final and is_final that means this text is accurate and it's a natural pause in the speakers speech. We need to send this to the assistant for processing
  if (transcriptionEvent.speech_final === true) {
    this.speechFinal = true; // this will prevent a utterance end which shows up after speechFinal from sending another response
    this.emit('transcription', this.finalResult);
    this.finalResult = '';
  } else {
    // if we receive a message without speechFinal reset speechFinal to false, this will allow any subsequent utteranceEnd messages to properly indicate the end of a message
    this.speechFinal = false;
  }
} else {
  this.emit('utterance', text);
}
});

you can get into a loop where the console.log output looks like this:

 is_final:  false speech_final:  false
 is_final:  false speech_final:  false
 is_final:  true speech_final:  false
 is_final:  false speech_final:  false
 is_final:  true speech_final:  true
 is_final:  false speech_final:  false

Where the speech_final is true but NO transcription event is emitted, even though there is content in the finalResult.

[jpage-bridge]

If there is a reason for this I would love to know! cause I removed the text.trim().length > 0 from the outer conditional and added it inside.

[batsector]

I am also noticing that the speech recognition keeps getting stuck and I have to say a random word for it to end an utterance. Did you find a fix?

[jpage-bridge]

The change I made was to move the text.trim().length > 0 from the first conditional. Sometimes the final string of transcribed text is an empty string. As long as is_final and speech_final are both true, you want to emit the finalResult.

if (transcriptionEvent.is_final === true) {
        if (text.trim().length > 0) {
          this.finalResult += ` ${text}`;
        }
        // if speech_final and is_final that means this text is accurate and it's a natural pause in the speakers speech. We need to send this to the assistant for processing
        if (transcriptionEvent.speech_final === true) {
          this.speechFinal = true; // this will prevent a utterance end which shows up after speechFinal from sending another response
          this.emit('transcription', this.finalResult);
          this.finalResult = '';
        } else if (text.trim().length == 0) {
          this.emit('utterance', text);
        } else {
          // If we receive a message without speechFinal reset speechFinal to false, this will allow any subsequent utteranceEnd messages to properly indicate the end of a message
          this.speechFinal = false;
        }
      } else {
        this.emit('utterance', text);
      }
});