whisper-client

Mic → Whisper → ChatGPT → Speech for React-Native

Lightweight drop-in that turns voice into AI answers in a single call. Works on bare React-Native 0.72 + or Expo config-plugin builds.

---

✨ Features

🔈 Record	📝 Transcribe	🤖 Chat	🔊 Speak
Saves temp .m4a using react-native-audio-recorder-player.	Sends to OpenAI Whisper (audio.transcriptions.create).	Streams to ChatGPT (any model, gpt-4o-mini default).	Replies via device TTS (RN-TTS) or OpenAI Audio TTS.

---

1 · Installation

npm install whisper-client
npx pod-install         # iOS pods

Installs & autolinks:

• react-native-audio-recorder-player
• react-native-permissions
• openai

---

2 · Permissions

Android – AndroidManifest.xml

<uses-permission android:name="android.permission.RECORD_AUDIO" />

iOS – Info.plist

<key>NSMicrophoneUsageDescription</key>
<string>This app needs your microphone for voice interviews.</string>

After editing Info.plist, run npx pod-install (or expo prebuild).

---

3 · Quick Start

import React, { useRef, useState } from 'react';
import { View, Button, Text } from 'react-native';
import { WhisperClient } from 'whisper-client';

export default function InterviewScreen() {
  const [speech, setSpeech] = useState('');
  const [reply,  setReply]  = useState('');

  // Keep one instance to preserve conversation history
  const vc = useRef(
    new WhisperClient(process.env.OPENAI_API_KEY!, {
      chatModel: 'gpt-4o-mini',   // optional override
      ttsEngine: 'device',        // 'device' | 'openai'
      language:  'en',
    }),
  ).current;

  return (
    <View style={{ flex: 1, gap: 12, padding: 24 }}>
      <Button title="Start Recording" onPress={vc.startRecording} />

      <Button
        title="Stop & Answer"
        onPress={async () => {
          const { transcript, answer } = await vc.stopAndAnswer();
          setSpeech(transcript);
          setReply(answer);
        }}
      />

      <Text style={{ marginTop: 16, fontWeight: '600' }}>You said:</Text>
      <Text>{speech}</Text>

      <Text style={{ marginTop: 16, fontWeight: '600' }}>AI replied:</Text>
      <Text>{reply}</Text>
    </View>
  );
}

---

4 · API

Constructor / Method	Purpose
new WhisperClient(apiKey, opts?)	Build a reusable instance. • opts.whisperModel default 'whisper-1' • opts.chatModel default 'gpt-4o-mini' • opts.language default 'en' • opts.ttsEngine 'device' | 'openai' (default 'device') • opts.systemPrompt custom system role • opts.onState(state) callback (idle → recording → transcribing → thinking → speaking)
startRecording()	Opens the mic and begins writing to a temp file.
stopAndAnswer() → { transcript, answer }	Stops recording, sends audio → Whisper → Chat → TTS, returns both strings.
nextQuestion() → { answer }	Ask ChatGPT without recording (e.g. next OSCE question).
cancel()	Abort any in-flight request or playback.
destroy()	Release native resources (call on unmount).

---

5 · Troubleshooting

Problem	Fix
Mic permission denied	Ensure runtime prompt accepted / Info.plist key present.
TS errors for AudioSet enums	Upgrade react-native-audio-recorder-player ≥ 3.6.
OpenAI 401 / network errors	Check OPENAI_API_KEY and connectivity.
Latency > 4 s	Lower opts.maxAudioMs, use Wi-Fi, or prefer device TTS.

---

6 · Roadmap

• Streaming partial transcripts & GPT tokens
• Silence detection → auto-stop recording
• Local transcript caching
• LangChain agent plug-in

---

7. Contributions

https://github.com/apium-io/whisper-client#

License

MIT © 2025 Apium Innovations