path: root/apps/web/src/voiceover/enhancements.ts

"use client";

export interface VoiceOverEnhancementSettings {
	voiceBoost: boolean;
	noiseGate: boolean;
	highPass: boolean;
	deEss: boolean;
	limiter: boolean;
	normalize: boolean;
}

export const DEFAULT_VOICE_OVER_ENHANCEMENTS: VoiceOverEnhancementSettings = {
	voiceBoost: true,
	noiseGate: true,
	highPass: true,
	deEss: false,
	limiter: true,
	normalize: true,
};

export const VOICE_OVER_ENHANCEMENT_LABELS: Record<
	keyof VoiceOverEnhancementSettings,
	{ label: string; description: string }
> = {
	voiceBoost: {
		label: "Voice boost",
		description: "Adds compression and presence for a stronger voice.",
	},
	noiseGate: {
		label: "Noise gate",
		description: "Turns down quiet room noise between phrases.",
	},
	highPass: {
		label: "High-pass",
		description: "Removes low rumble and desk vibration.",
	},
	deEss: {
		label: "De-ess",
		description: "Softens harsh S sounds and upper fizz.",
	},
	limiter: {
		label: "Limiter",
		description: "Catches peaks after boosts.",
	},
	normalize: {
		label: "Normalize",
		description: "Raises the finished take to a consistent level.",
	},
};

const WAV_BITS_PER_SAMPLE = 16;
const WAV_HEADER_BYTES = 44;
const TARGET_PEAK = 0.92;
const NOISE_GATE_THRESHOLD = 0.018;
const NOISE_GATE_FLOOR = 0.18;

export function hasEnabledVoiceOverEnhancement({
	settings,
}: {
	settings: VoiceOverEnhancementSettings;
}): boolean {
	return Object.values(settings).some(Boolean);
}

export async function applyVoiceOverEnhancementsToBlob({
	blob,
	settings,
}: {
	blob: Blob;
	settings: VoiceOverEnhancementSettings;
}): Promise<Blob> {
	if (!hasEnabledVoiceOverEnhancement({ settings })) {
		return blob;
	}

	const audioContext = createAudioContext();
	try {
		const buffer = await audioContext.decodeAudioData(await blob.arrayBuffer());
		const renderedBuffer = await renderEnhancedVoiceOverBuffer({
			buffer,
			settings,
		});
		return encodeAudioBufferToWav({ buffer: renderedBuffer });
	} finally {
		void audioContext.close();
	}
}

export async function renderEnhancedVoiceOverBuffer({
	buffer,
	settings,
}: {
	buffer: AudioBuffer;
	settings: VoiceOverEnhancementSettings;
}): Promise<AudioBuffer> {
	const offlineContext = new OfflineAudioContext(
		buffer.numberOfChannels,
		Math.max(1, buffer.length),
		buffer.sampleRate,
	);
	const source = offlineContext.createBufferSource();
	source.buffer = settings.noiseGate
		? applyNoiseGateToBuffer({
				buffer,
				context: offlineContext,
				threshold: NOISE_GATE_THRESHOLD,
				floor: NOISE_GATE_FLOOR,
			})
		: buffer;

	let current: AudioNode = source;

	if (settings.highPass) {
		const highPass = offlineContext.createBiquadFilter();
		highPass.type = "highpass";
		highPass.frequency.value = 85;
		highPass.Q.value = 0.7;
		current.connect(highPass);
		current = highPass;
	}

	if (settings.voiceBoost) {
		const presence = offlineContext.createBiquadFilter();
		presence.type = "peaking";
		presence.frequency.value = 3200;
		presence.Q.value = 0.9;
		presence.gain.value = 2.5;

		const compressor = offlineContext.createDynamicsCompressor();
		compressor.threshold.value = -24;
		compressor.knee.value = 12;
		compressor.ratio.value = 3.5;
		compressor.attack.value = 0.004;
		compressor.release.value = 0.16;

		const makeupGain = offlineContext.createGain();
		makeupGain.gain.value = 1.45;

		current.connect(presence);
		presence.connect(compressor);
		compressor.connect(makeupGain);
		current = makeupGain;
	}

	if (settings.deEss) {
		const deEss = offlineContext.createBiquadFilter();
		deEss.type = "peaking";
		deEss.frequency.value = 6400;
		deEss.Q.value = 2.2;
		deEss.gain.value = -3;
		current.connect(deEss);
		current = deEss;
	}

	if (settings.limiter) {
		const limiter = offlineContext.createDynamicsCompressor();
		limiter.threshold.value = -1;
		limiter.knee.value = 0;
		limiter.ratio.value = 20;
		limiter.attack.value = 0.001;
		limiter.release.value = 0.1;
		current.connect(limiter);
		current = limiter;
	}

	current.connect(offlineContext.destination);
	source.start(0);

	const rendered = await offlineContext.startRendering();
	if (settings.normalize) {
		normalizeAudioBufferPeak({ buffer: rendered, targetPeak: TARGET_PEAK });
	}
	clampAudioBuffer({ buffer: rendered, maxPeak: 0.98 });
	return rendered;
}

export function encodeAudioBufferToWav({
	buffer,
}: {
	buffer: AudioBuffer;
}): Blob {
	const numChannels = buffer.numberOfChannels;
	const bytesPerSample = WAV_BITS_PER_SAMPLE / 8;
	const dataSize = buffer.length * numChannels * bytesPerSample;
	const arrayBuffer = new ArrayBuffer(WAV_HEADER_BYTES + dataSize);
	const view = new DataView(arrayBuffer);

	writeAsciiString({ view, offset: 0, value: "RIFF" });
	view.setUint32(4, 36 + dataSize, true);
	writeAsciiString({ view, offset: 8, value: "WAVE" });
	writeAsciiString({ view, offset: 12, value: "fmt " });
	view.setUint32(16, 16, true);
	view.setUint16(20, 1, true);
	view.setUint16(22, numChannels, true);
	view.setUint32(24, buffer.sampleRate, true);
	view.setUint32(28, buffer.sampleRate * numChannels * bytesPerSample, true);
	view.setUint16(32, numChannels * bytesPerSample, true);
	view.setUint16(34, WAV_BITS_PER_SAMPLE, true);
	writeAsciiString({ view, offset: 36, value: "data" });
	view.setUint32(40, dataSize, true);

	let offset = WAV_HEADER_BYTES;
	for (let sampleIndex = 0; sampleIndex < buffer.length; sampleIndex++) {
		for (let channel = 0; channel < numChannels; channel++) {
			const sample = Math.max(
				-1,
				Math.min(1, buffer.getChannelData(channel)[sampleIndex] ?? 0),
			);
			const int16 = sample < 0 ? sample * 0x8000 : sample * 0x7fff;
			view.setInt16(offset, int16, true);
			offset += bytesPerSample;
		}
	}

	return new Blob([arrayBuffer], { type: "audio/wav" });
}

export function getAudioBufferPeak({ buffer }: { buffer: AudioBuffer }): number {
	let peak = 0;
	for (let channel = 0; channel < buffer.numberOfChannels; channel++) {
		const channelData = buffer.getChannelData(channel);
		for (let index = 0; index < channelData.length; index++) {
			peak = Math.max(peak, Math.abs(channelData[index] ?? 0));
		}
	}
	return peak;
}

function normalizeAudioBufferPeak({
	buffer,
	targetPeak,
}: {
	buffer: AudioBuffer;
	targetPeak: number;
}): void {
	const peak = getAudioBufferPeak({ buffer });
	if (peak <= 0 || peak >= targetPeak) {
		return;
	}

	const gain = targetPeak / peak;
	for (let channel = 0; channel < buffer.numberOfChannels; channel++) {
		const channelData = buffer.getChannelData(channel);
		for (let index = 0; index < channelData.length; index++) {
			channelData[index] *= gain;
		}
	}
}

function clampAudioBuffer({
	buffer,
	maxPeak,
}: {
	buffer: AudioBuffer;
	maxPeak: number;
}): void {
	for (let channel = 0; channel < buffer.numberOfChannels; channel++) {
		const channelData = buffer.getChannelData(channel);
		for (let index = 0; index < channelData.length; index++) {
			channelData[index] = Math.max(
				-maxPeak,
				Math.min(maxPeak, channelData[index] ?? 0),
			);
		}
	}
}

function applyNoiseGateToBuffer({
	buffer,
	context,
	threshold,
	floor,
}: {
	buffer: AudioBuffer;
	context: OfflineAudioContext;
	threshold: number;
	floor: number;
}): AudioBuffer {
	const gated = context.createBuffer(
		buffer.numberOfChannels,
		buffer.length,
		buffer.sampleRate,
	);

	for (let channel = 0; channel < buffer.numberOfChannels; channel++) {
		const input = buffer.getChannelData(channel);
		const output = gated.getChannelData(channel);
		for (let index = 0; index < input.length; index++) {
			const sample = input[index] ?? 0;
			output[index] =
				Math.abs(sample) < threshold
					? sample * floor
					: sample;
		}
	}

	return gated;
}

function createAudioContext(): AudioContext {
	const AudioContextConstructor =
		window.AudioContext ||
		(window as typeof window & { webkitAudioContext?: typeof AudioContext })
			.webkitAudioContext;

	return new AudioContextConstructor();
}

function writeAsciiString({
	view,
	offset,
	value,
}: {
	view: DataView;
	offset: number;
	value: string;
}): void {
	for (let index = 0; index < value.length; index++) {
		view.setUint8(offset + index, value.charCodeAt(index));
	}
}