Fix transcription sample-rate handling (#740)

Fix for #703.

Adds an optional sampleRate parameter to the decodeAudioToFloat32() function. The caption transcriber passes in a value of 16000 to prepare audio in the format Whisper expects.

apps/web/src/components/editor/panels/assets/views/captions.tsx

@@ -11,7 +11,10 @@ import { useState, useRef } from "react";
 import { extractTimelineAudio } from "@/lib/media/mediabunny";
 import { useEditor } from "@/hooks/use-editor";
 import { DEFAULT_TEXT_ELEMENT } from "@/constants/text-constants";
-import { TRANSCRIPTION_LANGUAGES } from "@/constants/transcription-constants";
+import {
+	DEFAULT_TRANSCRIPTION_SAMPLE_RATE,
+	TRANSCRIPTION_LANGUAGES,
+} from "@/constants/transcription-constants";
 import type {
 	TranscriptionLanguage,
 	TranscriptionProgress,
@@ -52,7 +55,10 @@ export function Captions() {
 			});
 
 			setProcessingStep("Preparing audio...");
-			const { samples } = await decodeAudioToFloat32({ audioBlob });
+			const { samples } = await decodeAudioToFloat32({
+				audioBlob,
+				sampleRate: DEFAULT_TRANSCRIPTION_SAMPLE_RATE,
+			});
 
 			const result = await transcriptionService.transcribe({
 				audioData: samples,

apps/web/src/constants/transcription-constants.ts

@@ -51,6 +51,8 @@ export const TRANSCRIPTION_MODELS: TranscriptionModel[] = [
 export const DEFAULT_TRANSCRIPTION_MODEL: TranscriptionModelId =
 	"whisper-small";
 
+export const DEFAULT_TRANSCRIPTION_SAMPLE_RATE = 16000;
+
 export const DEFAULT_CHUNK_LENGTH_SECONDS = 30;
 export const DEFAULT_STRIDE_SECONDS = 5;
 

apps/web/src/lib/media/audio.ts

@@ -34,10 +34,12 @@ export interface DecodedAudio {
 
 export async function decodeAudioToFloat32({
 	audioBlob,
+	sampleRate,
 }: {
 	audioBlob: Blob;
+	sampleRate?: number;
 }): Promise<DecodedAudio> {
-	const audioContext = createAudioContext();
+	const audioContext = createAudioContext({ sampleRate });
 	const arrayBuffer = await audioBlob.arrayBuffer();
 	const audioBuffer = await audioContext.decodeAudioData(arrayBuffer);