Enhanced ChatGPT Clone: Features Agents, MCP, Skills, DeepSeek, Anthropic, AWS, OpenAI, Responses API, Azure, Groq, o1, GPT-5, Mistral, OpenRouter, Vertex AI, Gemini, Artifacts, AI model switching, message search, Code Interpreter, langchain, DALL-E-3, OpenAPI Actions, Functions, Secure Multi-User Auth, Presets, open-source for self-hosting. Active
TypeScript aianthropicartifactsawsazurechatgptchatgpt-cloneclaudeclonedeepseekgeminigooglegpt-5librechatmcpo1openairesponses-apivisionwebui
import { useRef, useEffect, useCallback } from 'react';
/** Loudness that draws an empty bar, and loudness that draws a full one. Room
* noise sits near the floor and ordinary speech lands mid-range, which a
* linear scale could not do: speech is a few percent of full scale. */
const FLOOR_DB = -55;
const CEILING_DB = -12;
/**
* The visualiser's microphone is left unprocessed. Automatic gain starts low
* and ramps over the first seconds, which drew the opening words as a flat
* line, and noise suppression holds back speech while it learns the room. The
* recording sent for transcription opens its own stream with its own settings.
*/
const RAW_AUDIO: MediaTrackConstraints = {
autoGainControl: false,
noiseSuppression: false,
echoCancellation: false,
};
/**
* Loudness of one time-domain frame on a 0-1 scale. RMS around the 128
* midpoint, then decibels, because hearing is logarithmic and a linear mapping
* left speech drawn as a row of dots.
*/
export function frameLevel(buffer: Uint8Array): number {
if (buffer.length === 0) {
return 0;
}
let sum = 0;
for (let i = 0; i < buffer.length; i++) {
const deviation = (buffer[i] - 128) / 128;
sum += deviation * deviation;
}
const rms = Math.sqrt(sum / buffer.length);
const db = 20 * Math.log10(Math.max(rms, 1e-6));
return Math.min(1, Math.max(0, (db - FLOOR_DB) / (CEILING_DB - FLOOR_DB)));
}
interface Tap {
analyser: AnalyserNode;
buffer: Uint8Array<ArrayBuffer>;
}
/**
* The microphone's current loudness, read on demand. Returns a reader rather
* than a stream of values so the caller samples on its own animation frame and
* nothing re-renders per sample; the reader answers `null` until the
* microphone is open.
*
* Deliberately opens its own stream rather than reaching into the speech hooks:
* the browser and external transcription paths are completely different (the
* Web Speech API never exposes audio at all), and a visualiser has no business
* interfering with either. The stream is torn down the moment `active` goes
* false, so the recording indicator does not linger.
*/
export default function useAudioLevels(active: boolean): () => number | null {
const tapRef = useRef<Tap | null>(null);
useEffect(() => {
if (!active) {
return;
}
let stream: MediaStream | undefined;
let context: AudioContext | undefined;
let cancelled = false;
const start = async () => {
try {
stream = await navigator.mediaDevices.getUserMedia({ audio: RAW_AUDIO });
} catch {
/* Permission denied or no device: the trace still draws, just flat. */
return;
}
if (cancelled) {
stream.getTracks().forEach((track) => track.stop());
return;
}
context = new AudioContext();
/* Created after the permission prompt resolved, which is outside the click
that started recording, so the browser may hand it over suspended; a
suspended graph reads as silence forever. */
if (context.state === 'suspended') {
void context.resume();
}
const analyser = context.createAnalyser();
analyser.fftSize = 1024;
context.createMediaStreamSource(stream).connect(analyser);
tapRef.current = { analyser, buffer: new Uint8Array(analyser.fftSize) };
};
void start();
return () => {
cancelled = true;
tapRef.current = null;
stream?.getTracks().forEach((track) => track.stop());
void context?.close();
};
}, [active]);
return useCallback(() => {
const tap = tapRef.current;
if (tap == null) {
return null;
}
tap.analyser.getByteTimeDomainData(tap.buffer);
return frameLevel(tap.buffer);
}, []);
}