fix: update whisper models and split-on-word flag

README.md

@@ -94,15 +94,35 @@ await nodewhisper(filePath, {
 const MODELS_LIST = [
 	'tiny',
 	'tiny.en',
+	'tiny-q5_1',
+	'tiny.en-q5_1',
+	'tiny-q8_0',
 	'base',
 	'base.en',
+	'base-q5_1',
+	'base.en-q5_1',
+	'base-q8_0',
 	'small',
 	'small.en',
+	'small.en-tdrz',
+	'small-q5_1',
+	'small.en-q5_1',
+	'small-q8_0',
 	'medium',
 	'medium.en',
+	'medium-q5_0',
+	'medium.en-q5_0',
+	'medium-q8_0',
 	'large-v1',
-	'large',
+	'large-v2',
+	'large-v2-q5_0',
+	'large-v2-q8_0',
+	'large-v3',
+	'large-v3-q5_0',
 	'large-v3-turbo',
+	'large-v3-turbo-q5_0',
+	'large-v3-turbo-q8_0',
+	'large', // backward-compatible alias for large-v3
 ]
 ```
 

src/WhisperHelper.ts

@@ -1,6 +1,6 @@
 import path from 'path'
 import fs from 'fs'
-import { MODELS_LIST, MODEL_OBJECT, WHISPER_CPP_PATH } from './constants'
+import { isModelName, LEGACY_MODEL_FILES, MODELS_LIST, MODEL_OBJECT, WHISPER_CPP_PATH } from './constants'
 import { IOptions } from '.'
 
 // Get the correct executable path based on platform and build system
@@ -26,35 +26,30 @@ function getExecutablePath(): string {
 }
 
 export const constructCommand = (filePath: string, args: IOptions): string => {
-	let errors: string[] = []
-	const modelName = MODEL_OBJECT[args.modelName as keyof typeof MODEL_OBJECT]
-
 	if (!args.modelName) {
-		errors.push('[Nodejs-whisper] Error: Provide model name')
+		throw new Error('[Nodejs-whisper] Error: Provide model name')
 	}
 
-	if (!MODELS_LIST.includes(args.modelName)) {
-		errors.push(`[Nodejs-whisper] Error: Enter a valid model name. Available models are: ${MODELS_LIST.join(', ')}`)
+	if (!isModelName(args.modelName)) {
+		throw new Error(
+			`[Nodejs-whisper] Error: Enter a valid model name. Available models are: ${MODELS_LIST.join(', ')}`
+		)
 	}
 
-	if (errors.length > 0) {
-		throw new Error(errors.join('\n'))
-	}
+	const modelName = MODEL_OBJECT[args.modelName]
 
-	const modelPath = args.modelRootPath
-		? path.resolve(args.modelRootPath, modelName)
-		: path.join(WHISPER_CPP_PATH, 'models', modelName)
+	const modelDirectory = args.modelRootPath ? path.resolve(args.modelRootPath) : path.join(WHISPER_CPP_PATH, 'models')
+	const modelFileNames = [modelName, ...(LEGACY_MODEL_FILES[args.modelName] || [])]
+	const modelFileName =
+		modelFileNames.find(fileName => fs.existsSync(path.join(modelDirectory, fileName))) || modelName
+	const modelPath = path.join(modelDirectory, modelFileName)
 
 	if (!fs.existsSync(modelPath)) {
-		errors.push(
+		throw new Error(
 			`[Nodejs-whisper] Error: Model file does not exist at ${modelPath}. Please ensure the model is downloaded and correctly placed.`
 		)
 	}
 
-	if (errors.length > 0) {
-		throw new Error(errors.join('\n'))
-	}
-
 	// Get the actual executable path
 	const executablePath = getExecutablePath()
 	if (!executablePath) {
@@ -69,7 +64,7 @@ export const constructCommand = (filePath: string, args: IOptions): string => {
 		return `"${arg}"`
 	}
 
-	const modelArg = args.modelRootPath ? modelPath : `./models/${modelName}`
+	const modelArg = args.modelRootPath ? modelPath : `./models/${modelFileName}`
 
 	let command = `${escapeArg(executablePath)} ${constructOptionsFlags(args)} -l ${args.whisperOptions?.language || 'auto'} -m ${escapeArg(modelArg)} -f ${escapeArg(filePath)}`
 
@@ -89,7 +84,7 @@ const constructOptionsFlags = (args: IOptions): string => {
 		args.whisperOptions?.translateToEnglish ? '-tr ' : '',
 		args.whisperOptions?.wordTimestamps ? '-ml 1 ' : '',
 		args.whisperOptions?.timestamps_length ? `-ml ${args.whisperOptions.timestamps_length} ` : '',
-		args.whisperOptions?.splitOnWord ? '-sow true ' : '',
+		args.whisperOptions?.splitOnWord ? '-sow ' : '',
 		args.whisperOptions?.noGpu ? '-ng ' : '',
 	].join('')
 

src/autoDownloadModel.ts

@@ -1,7 +1,7 @@
 import path from 'path'
 import shell from 'shelljs'
 import fs from 'fs'
-import { MODEL_OBJECT, MODELS_LIST, WHISPER_CPP_PATH } from './constants'
+import { DOWNLOAD_MODEL_ALIASES, isModelName, LEGACY_MODEL_FILES, MODEL_OBJECT, WHISPER_CPP_PATH } from './constants'
 import { Logger } from './types'
 import { getCmakeConfigureCommand } from './buildConfig'
 
@@ -17,7 +17,7 @@ export default async function autoDownloadModel(
 		throw new Error('[Nodejs-whisper] Error: Model name must be provided.')
 	}
 
-	if (!MODELS_LIST.includes(autoDownloadModelName)) {
+	if (!isModelName(autoDownloadModelName)) {
 		throw new Error('[Nodejs-whisper] Error: Provide a valid model name')
 	}
 
@@ -27,14 +27,16 @@ export default async function autoDownloadModel(
 
 		fs.mkdirSync(downloadDirectory, { recursive: true })
 		shell.cd(modelDirectory)
-		const modelAlreadyExist = fs.existsSync(path.join(downloadDirectory, MODEL_OBJECT[autoDownloadModelName]))
+		const modelFiles = [MODEL_OBJECT[autoDownloadModelName], ...(LEGACY_MODEL_FILES[autoDownloadModelName] || [])]
+		const modelAlreadyExist = modelFiles.some(modelFile => fs.existsSync(path.join(downloadDirectory, modelFile)))
 
 		if (modelAlreadyExist) {
 			logger.debug(`[Nodejs-whisper] ${autoDownloadModelName} already exist. Skipping download.`)
 			return 'Models already exist. Skipping download.'
 		}
 
 		logger.debug(`[Nodejs-whisper] Auto-download Model: ${autoDownloadModelName}`)
+		const downloadModelName = DOWNLOAD_MODEL_ALIASES[autoDownloadModelName] || autoDownloadModelName
 
 		let scriptPath = './download-ggml-model.sh'
 		if (process.platform === 'win32') {
@@ -43,7 +45,7 @@ export default async function autoDownloadModel(
 
 		shell.chmod('+x', scriptPath)
 		const downloadPathArg = modelRootPath ? ` ${quoteShellArg(downloadDirectory)}` : ''
-		const result = shell.exec(`${scriptPath} ${autoDownloadModelName}${downloadPathArg}`)
+		const result = shell.exec(`${scriptPath} ${downloadModelName}${downloadPathArg}`)
 
 		if (result.code !== 0) {
 			throw new Error(`[Nodejs-whisper] Failed to download model: ${result.stderr}`)

src/constants.ts

@@ -1,47 +1,55 @@
 import path from 'path'
 
-export const MODELS_LIST = [
-	'tiny',
-	'tiny.en',
-	'base',
-	'base.en',
-	'small',
-	'small.en',
-	'medium',
-	'medium.en',
-	'large-v1',
-	'large',
-	'large-v3-turbo',
-]
-
-export const MODELS = [
-	'ggml-tiny.en.bin',
-	'ggml-tiny.bin',
-	'ggml-base.en.bin',
-	'ggml-base.bin',
-	'ggml-small.en.bin',
-	'ggml-small.bin',
-	'ggml-medium.en.bin',
-	'ggml-medium.bin',
-	'ggml-large-v1.bin',
-	'ggml-large.bin',
-	'ggml-large-v3-turbo.bin',
-]
-
 export const MODEL_OBJECT = {
 	tiny: 'ggml-tiny.bin',
 	'tiny.en': 'ggml-tiny.en.bin',
+	'tiny-q5_1': 'ggml-tiny-q5_1.bin',
+	'tiny.en-q5_1': 'ggml-tiny.en-q5_1.bin',
+	'tiny-q8_0': 'ggml-tiny-q8_0.bin',
 	base: 'ggml-base.bin',
 	'base.en': 'ggml-base.en.bin',
+	'base-q5_1': 'ggml-base-q5_1.bin',
+	'base.en-q5_1': 'ggml-base.en-q5_1.bin',
+	'base-q8_0': 'ggml-base-q8_0.bin',
 	small: 'ggml-small.bin',
 	'small.en': 'ggml-small.en.bin',
+	'small.en-tdrz': 'ggml-small.en-tdrz.bin',
+	'small-q5_1': 'ggml-small-q5_1.bin',
+	'small.en-q5_1': 'ggml-small.en-q5_1.bin',
+	'small-q8_0': 'ggml-small-q8_0.bin',
 	medium: 'ggml-medium.bin',
 	'medium.en': 'ggml-medium.en.bin',
+	'medium-q5_0': 'ggml-medium-q5_0.bin',
+	'medium.en-q5_0': 'ggml-medium.en-q5_0.bin',
+	'medium-q8_0': 'ggml-medium-q8_0.bin',
 	'large-v1': 'ggml-large-v1.bin',
-	large: 'ggml-large.bin',
+	'large-v2': 'ggml-large-v2.bin',
+	'large-v2-q5_0': 'ggml-large-v2-q5_0.bin',
+	'large-v2-q8_0': 'ggml-large-v2-q8_0.bin',
+	'large-v3': 'ggml-large-v3.bin',
+	'large-v3-q5_0': 'ggml-large-v3-q5_0.bin',
 	'large-v3-turbo': 'ggml-large-v3-turbo.bin',
+	'large-v3-turbo-q5_0': 'ggml-large-v3-turbo-q5_0.bin',
+	'large-v3-turbo-q8_0': 'ggml-large-v3-turbo-q8_0.bin',
+	large: 'ggml-large-v3.bin',
+}
+
+export type ModelName = keyof typeof MODEL_OBJECT
+
+export const MODELS_LIST = Object.keys(MODEL_OBJECT) as ModelName[]
+export const MODELS = Object.values(MODEL_OBJECT)
+
+export const DOWNLOAD_MODEL_ALIASES: Partial<Record<ModelName, ModelName>> = {
+	large: 'large-v3',
+}
+
+export const LEGACY_MODEL_FILES: Partial<Record<ModelName, string[]>> = {
+	large: ['ggml-large.bin'],
 }
 
+export const isModelName = (modelName: unknown): modelName is ModelName =>
+	typeof modelName === 'string' && Object.prototype.hasOwnProperty.call(MODEL_OBJECT, modelName)
+
 export const DEFAULT_MODEL = 'tiny.en'
 
 export const WHISPER_CPP_PATH = path.join(__dirname, '..', 'cpp', 'whisper.cpp')

src/downloadModel.ts

@@ -5,11 +5,20 @@
 import path from 'path'
 import shell from 'shelljs'
 import readlineSync from 'readline-sync'
-import { MODELS_LIST, DEFAULT_MODEL, MODELS, WHISPER_CPP_PATH, MODEL_OBJECT } from './constants'
+import {
+	DEFAULT_MODEL,
+	DOWNLOAD_MODEL_ALIASES,
+	isModelName,
+	LEGACY_MODEL_FILES,
+	ModelName,
+	MODELS_LIST,
+	WHISPER_CPP_PATH,
+	MODEL_OBJECT,
+} from './constants'
 import fs from 'fs'
 import { Logger } from './types'
 import { getCmakeConfigureCommand } from './buildConfig'
-const askForModel = async (logger: Logger = console): Promise<string> => {
+const askForModel = async (logger: Logger = console): Promise<ModelName> => {
 	const answer = await readlineSync.question(
 		`\n[Nodejs-whisper] Enter model name (e.g. 'tiny.en') or 'cancel' to exit\n(ENTER for tiny.en): `
 	)
@@ -22,7 +31,7 @@ const askForModel = async (logger: Logger = console): Promise<string> => {
 	else if (answer === '') {
 		logger.log('[Nodejs-whisper] Going with', DEFAULT_MODEL)
 		return DEFAULT_MODEL
-	} else if (!MODELS_LIST.includes(answer)) {
+	} else if (!isModelName(answer)) {
 		logger.log(
 			'\n[Nodejs-whisper] FAIL: Name not found. Check your spelling OR quit wizard and use custom model.\n'
 		)
@@ -53,11 +62,11 @@ async function downloadModel(logger: Logger = console) {
 	try {
 		shell.cd(path.join(WHISPER_CPP_PATH, 'models'))
 
-		let anyModelExist = []
+		const anyModelExist: ModelName[] = []
 
 		MODELS_LIST.forEach(model => {
-			if (!fs.existsSync(path.join(WHISPER_CPP_PATH, 'models', MODEL_OBJECT[model]))) {
-			} else {
+			const modelFiles = [MODEL_OBJECT[model], ...(LEGACY_MODEL_FILES[model] || [])]
+			if (modelFiles.some(modelFile => fs.existsSync(path.join(WHISPER_CPP_PATH, 'models', modelFile)))) {
 				anyModelExist.push(model)
 			}
 		})
@@ -68,21 +77,7 @@ async function downloadModel(logger: Logger = console) {
 			logger.log('\n[Nodejs-whisper] You can install additional models from the list below.\n')
 		}
 
-		logger.log(`
-| Model          | Disk   | RAM     |
-|----------------|--------|---------|
-| tiny           |  75 MB | ~390 MB |
-| tiny.en        |  75 MB | ~390 MB |
-| base           | 142 MB | ~500 MB |
-| base.en        | 142 MB | ~500 MB |
-| small          | 466 MB | ~1.0 GB |
-| small.en       | 466 MB | ~1.0 GB |
-| medium         | 1.5 GB | ~2.6 GB |
-| medium.en      | 1.5 GB | ~2.6 GB |
-| large-v1       | 2.9 GB | ~4.7 GB |
-| large          | 2.9 GB | ~4.7 GB |
-| large-v3-turbo | 1.5 GB | ~2.6 GB |
-`)
+		logger.log(`[Nodejs-whisper] Available models:\n${MODELS_LIST.join('\n')}`)
 
 		const downloaderScript = process.platform === 'win32' ? 'download-ggml-model.cmd' : './download-ggml-model.sh'
 
@@ -91,11 +86,12 @@ async function downloadModel(logger: Logger = console) {
 		}
 
 		const modelName = await askForModel()
+		const downloadModelName = DOWNLOAD_MODEL_ALIASES[modelName] || modelName
 
 		const scriptPath = downloaderScript
 
 		shell.chmod('+x', scriptPath)
-		shell.exec(`${scriptPath} ${modelName}`)
+		shell.exec(`${scriptPath} ${downloadModelName}`)
 
 		logger.log('[Nodejs-whisper] Attempting to build whisper.cpp...\n')
 		shell.cd('../')

test/core.integration.js

@@ -48,6 +48,7 @@ test(
 				whisperOptions: {
 					noGpu: true,
 					outputInVtt: true,
+					splitOnWord: true,
 					timestamps_length: 14,
 				},
 			})
@@ -71,6 +72,8 @@ test(
 		assert.match(normalizedTranscript, /i want your.*forgiveness/)
 		assert.match(loggedTranscript.toLowerCase(), /i do not want/)
 		assert.match(debugOutput, /whisper_init_with_params_no_state:/)
+		assert.match(debugOutput, /-sow(?:\s|$)/)
+		assert.doesNotMatch(debugOutput, /-sow\s+true/)
 		assert.deepEqual(leakedWhisperOutput, [])
 
 		assert.equal(fs.existsSync(outputFile), true, 'VTT output should be created')