​
Overview

Trieve provides a simple way to integrate voice search into your application. This guide will walk you through the steps to set up voice search with Trieve.

For a full implementation example, take a look at the way we implement voice search in our search component.

​
Capturing Audio from the Microphone

To enable voice search, you’ll need to capture audio from the user’s microphone using the browser’s MediaRecorder API.

​
1. Create a Voice Search Button

The following React component lets users start and stop voice recording:

import React, { useState } from "react";

export const VoiceSearchButton = () => {
  const [recording, setRecording] = useState(false);
  const [mediaRecorder, setMediaRecorder] = useState<MediaRecorder | null>(null);

  const startRecording = async () => {
    try {
      const stream = await navigator.mediaDevices.getUserMedia({ audio: true });
      const mimeType = navigator.userAgent.includes('Firefox') ? 'audio/webm' : 'audio/mp4';
      
      const recorder = new MediaRecorder(stream, { mimeType });
      let audioChunks: Blob[] = [];
      
      recorder.ondataavailable = (e) => audioChunks.push(e.data);
      recorder.onstop = async () => {
        const audioBlob = new Blob(audioChunks);
        const base64Audio = await convertBlobToBase64(audioBlob);
        handleSearch(base64Audio);
      };

      setMediaRecorder(recorder);
      recorder.start();
      setRecording(true);
    } catch (error) {
      console.error("Microphone access error:", error);
    }
  };

  const stopRecording = () => {
    mediaRecorder?.stop();
    setRecording(false);
  };

  return (
    <button
      onClick={recording ? stopRecording : startRecording}
      aria-label={recording ? "Stop recording" : "Start voice search"}
    >
      {recording ? <StopIcon /> : <MicIcon />}
    </button>
  );
};

​
Converting Audio to Base64

To send the recorded audio to Trieve, convert the audio blob into a base64 string.

const convertBlobToBase64 = (blob: Blob): Promise<string> => {
  return new Promise((resolve) => {
    const reader = new FileReader();
    reader.onloadend = () => {
      const base64String = (reader.result as string).split(',')[1];
      resolve(base64String);
    };
    reader.readAsDataURL(blob);
  });
};

Once the audio is recorded and converted to base64, send it to Trieve. The platform will transcribe the speech using OpenAI Whisper and return the search results based on the transcription.

We return the transcribed text in the response header as x-tr-query, which can be used to update the UI with the search query.

const handleSearch = async (audioBase64?: string) => {
  try {
    const response = await trieveClient.search({
      audio_base64: audioBase64,
      search_type: "hybrid",
      score_threshold: 0.5,
      page_size: 10
    });

    // Retrieve transcribed text from the response header
    const queryText = response.headers.get('x-tr-query');
    
    // Update UI with search results
    setSearchResults(response.data.chunks);
    setQuery(queryText || "");
  } catch (error) {
    console.error("Voice search error:", error);
  }
};

​
Best Practices

​
Optimize Performance

  • Keep recordings under 30 seconds to improve speed and accuracy.
  • Use high-quality microphones for clearer transcription results.

​
Ensure Browser Compatibility

Different browsers support different audio formats. Handle this by selecting the correct MIME type:

const isFirefox = navigator.userAgent.includes('Firefox');
const mimeType = isFirefox ? 'audio/webm' : 'audio/mp4';