eprec 1.10.2 → 1.12.0

package/src/cli.ts

@@ -27,6 +27,7 @@ import {
 	PromptCancelled,
 	createInquirerPrompter,
 	createPathPicker,
+	createStepProgressReporter,
 	isInteractive,
 	pauseActiveSpinner,
 	resumeActiveSpinner,
@@ -156,18 +157,30 @@ async function main(rawArgs = hideBin(process.argv)) {
 					}),
 			async (argv) => {
 				const transcribeArgs = await resolveTranscribeArgs(argv, context)
+				const progress = context.interactive
+					? createStepProgressReporter({ action: 'Transcribing audio' })
+					: undefined
 				let resultText = ''
 				await withSpinner(
 					'Transcribing audio',
 					async () => {
-						const result = await transcribeAudio(transcribeArgs.inputPath, {
-							modelPath: transcribeArgs.modelPath,
-							language: transcribeArgs.language,
-							threads: transcribeArgs.threads,
-							binaryPath: transcribeArgs.binaryPath,
-							outputBasePath: transcribeArgs.outputBasePath,
+						setLogHooks({
+							beforeLog: pauseActiveSpinner,
+							afterLog: resumeActiveSpinner,
 						})
-						resultText = result.text
+						try {
+							const result = await transcribeAudio(transcribeArgs.inputPath, {
+								modelPath: transcribeArgs.modelPath,
+								language: transcribeArgs.language,
+								threads: transcribeArgs.threads,
+								binaryPath: transcribeArgs.binaryPath,
+								outputBasePath: transcribeArgs.outputBasePath,
+								progress,
+							})
+							resultText = result.text
+						} finally {
+							setLogHooks({})
+						}
 					},
 					{
 						successText: 'Transcription complete',
@@ -203,16 +216,28 @@ async function main(rawArgs = hideBin(process.argv)) {
 					argv,
 					context,
 				)
+				const progress = context.interactive
+					? createStepProgressReporter({ action: 'Detecting speech' })
+					: undefined
 				let segments: unknown = []
 				await withSpinner(
 					'Detecting speech',
 					async () => {
-						await ensureFfmpegAvailable()
-						segments = await detectSpeechSegmentsForFile({
-							inputPath,
-							start,
-							end,
+						setLogHooks({
+							beforeLog: pauseActiveSpinner,
+							afterLog: resumeActiveSpinner,
 						})
+						try {
+							await ensureFfmpegAvailable()
+							segments = await detectSpeechSegmentsForFile({
+								inputPath,
+								start,
+								end,
+								progress,
+							})
+						} finally {
+							setLogHooks({})
+						}
 					},
 					{
 						successText: 'Speech detection complete',

package/src/speech-detection.ts

@@ -6,6 +6,7 @@ import { CONFIG } from '../process-course/config'
 import { formatSeconds, getMediaDurationSeconds } from './utils'
 import { speechFallback } from '../process-course/utils/audio-analysis'
 import type { SpeechBounds } from '../process-course/types'
+import type { StepProgressReporter } from '../progress-reporter'
 
 export type VadConfig = {
 	vadWindowSamples: number
@@ -27,6 +28,7 @@ export async function detectSpeechSegmentsWithVad(
 	samples: Float32Array,
 	sampleRate: number,
 	config: VadConfig,
+	options?: { onProgress?: () => void; updateStride?: number },
 ): Promise<VadSegment[]> {
 	const vadSession = await getVadSession(config)
 	const probabilities = await getVadProbabilities(
@@ -34,6 +36,7 @@ export async function detectSpeechSegmentsWithVad(
 		sampleRate,
 		config,
 		vadSession,
+		options,
 	)
 	return probabilitiesToSegments(
 		samples.length,
@@ -47,7 +50,10 @@ export async function detectSpeechSegmentsForFile(options: {
 	inputPath: string
 	start?: number
 	end?: number
+	progress?: StepProgressReporter
 }): Promise<SpeechSegment[]> {
+	const progress = options.progress
+	progress?.start({ stepCount: 1, label: 'Loading audio' })
 	const start = options.start ?? 0
 	if (!Number.isFinite(start) || start < 0) {
 		throw new Error('Start time must be a non-negative number.')
@@ -66,13 +72,31 @@ export async function detectSpeechSegmentsForFile(options: {
 		sampleRate: CONFIG.vadSampleRate,
 	})
 	if (samples.length === 0) {
+		progress?.finish('No audio')
 		return []
 	}
+	const windowSamples = CONFIG.vadWindowSamples
+	const totalWindows = Math.ceil(samples.length / windowSamples)
+	const updateStride = Math.max(1, Math.floor(totalWindows / 50))
+	const updateCount = Math.max(1, Math.ceil(totalWindows / updateStride))
+	progress?.start({ stepCount: updateCount, label: 'Running VAD' })
+	let progressUpdates = 0
 	const segments = await detectSpeechSegmentsWithVad(
 		samples,
 		CONFIG.vadSampleRate,
 		CONFIG,
+		{
+			onProgress: () => {
+				progressUpdates += 1
+				if (progressUpdates <= updateCount) {
+					progress?.step('Running VAD')
+				}
+			},
+			updateStride,
+		},
 	)
+	progress?.setLabel('Building segments')
+	progress?.finish('Complete')
 	return segments.map((segment) => ({
 		start: segment.start + start,
 		end: segment.end + start,
@@ -116,6 +140,7 @@ async function getVadProbabilities(
 	sampleRate: number,
 	config: VadConfig,
 	session: ort.InferenceSession,
+	options?: { onProgress?: () => void; updateStride?: number },
 ) {
 	const windowSamples = config.vadWindowSamples
 	const srTensor = new ort.Tensor(
@@ -126,6 +151,8 @@ async function getVadProbabilities(
 	const probabilities: number[] = []
 	let stateH = new Float32Array(2 * 1 * 64)
 	let stateC = new Float32Array(2 * 1 * 64)
+	const updateStride = Math.max(1, Math.floor(options?.updateStride ?? 1))
+	let updateIndex = 0
 
 	for (let offset = 0; offset < samples.length; offset += windowSamples) {
 		const chunk = samples.subarray(offset, offset + windowSamples)
@@ -154,6 +181,10 @@ async function getVadProbabilities(
 		probabilities.push((probTensor.data as Float32Array)[0] ?? 0)
 		stateH = new Float32Array(nextH.data as Float32Array)
 		stateC = new Float32Array(nextC.data as Float32Array)
+		if (updateIndex % updateStride === 0) {
+			options?.onProgress?.()
+		}
+		updateIndex += 1
 	}
 
 	return probabilities

package/src/whispercpp-transcribe.ts

@@ -1,6 +1,7 @@
 import path from 'node:path'
 import { mkdir } from 'node:fs/promises'
 import { runCommand } from './utils'
+import type { StepProgressReporter } from '../progress-reporter'
 
 const DEFAULT_MODEL_FILENAME = 'ggml-small.en.bin'
 const DEFAULT_MODEL_URL =
@@ -14,6 +15,7 @@ type TranscribeOptions = {
 	threads?: number
 	binaryPath?: string
 	outputBasePath?: string
+	progress?: StepProgressReporter
 }
 
 export type TranscriptSegment = {
@@ -36,6 +38,7 @@ export async function transcribeAudio(
 	audioPath: string,
 	options: TranscribeOptions = {},
 ): Promise<TranscriptionResult> {
+	const progress = options.progress
 	const resolvedAudioPath = path.resolve(audioPath)
 	const resolvedModelPath = path.resolve(
 		options.modelPath ?? getDefaultWhisperModelPath(),
@@ -49,7 +52,9 @@ export async function transcribeAudio(
 			`${path.parse(resolvedAudioPath).name}-transcript`,
 		)
 
-	await ensureModelFile(resolvedModelPath)
+	const totalSteps = 3
+	progress?.start({ stepCount: totalSteps, label: 'Checking model' })
+	await ensureModelFile(resolvedModelPath, progress)
 
 	const args = [
 		binaryPath,
@@ -69,17 +74,23 @@ export async function transcribeAudio(
 		args.push('-t', String(options.threads))
 	}
 
+	progress?.step('Transcribing audio')
 	const result = await runCommand(args)
+	progress?.step('Reading output')
 	const transcriptPath = `${outputBasePath}.txt`
 	const transcript = await readTranscriptText(transcriptPath, result.stdout)
 	const { segments, source } = await readTranscriptSegments(
 		`${outputBasePath}.json`,
 	)
 	const normalized = normalizeTranscriptText(transcript)
+	progress?.finish('Complete')
 	return { text: normalized, segments, segmentsSource: source }
 }
 
-async function ensureModelFile(modelPath: string) {
+async function ensureModelFile(
+	modelPath: string,
+	progress?: StepProgressReporter,
+) {
 	const file = Bun.file(modelPath)
 	if (await file.exists()) {
 		return
@@ -90,6 +101,7 @@ async function ensureModelFile(modelPath: string) {
 		throw new Error(`Whisper model not found at ${modelPath}.`)
 	}
 
+	progress?.setLabel('Downloading model')
 	await mkdir(path.dirname(modelPath), { recursive: true })
 	const response = await fetch(DEFAULT_MODEL_URL)
 	if (!response.ok) {