echogarden 2.7.0 → 2.8.0

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "echogarden",
-	"version": "2.7.0",
+	"version": "2.8.0",
 	"description": "An easy-to-use speech toolset. Includes tools for synthesis, recognition, alignment, speech translation, language detection, source separation and more.",
 	"author": "Rotem Dan",
 	"license": "GPL-3.0",
@@ -52,27 +52,27 @@
 		"echogarden": "./dist/cli/CLILauncher.js"
 	},
 	"dependencies": {
-		"@aws-sdk/client-polly": "^3.799.0",
-		"@aws-sdk/client-transcribe-streaming": "^3.799.0",
-		"@echogarden/audio-io": "^0.3.0",
+		"@aws-sdk/client-polly": "^3.810.0",
+		"@aws-sdk/client-transcribe-streaming": "^3.810.0",
+		"@echogarden/audio-io": "^0.4.1",
 		"@echogarden/espeak-ng-emscripten": "^0.3.3",
 		"@echogarden/fasttext-wasm": "^0.1.0",
 		"@echogarden/flite-wasi": "^0.1.1",
 		"@echogarden/fvad-wasm": "^0.2.0",
 		"@echogarden/gnuspeech-wasm": "^0.2.1",
+		"@echogarden/icu-segmentation-wasm": "^0.2.1",
 		"@echogarden/pffft-wasm": "^0.4.2",
 		"@echogarden/rnnoise-wasm": "^0.2.0",
 		"@echogarden/rubberband-wasm": "^0.2.0",
 		"@echogarden/sonic-wasm": "^0.2.0",
 		"@echogarden/speex-resampler-wasm": "^0.3.0",
 		"@echogarden/svoxpico-wasm": "^0.2.0",
-		"@echogarden/transformers-nodejs-lite": "^2.17.1-lite.3",
-		"@echogarden/wave-codec": "^0.3.2",
+		"@echogarden/text-segmentation": "^0.3.1",
+		"@echogarden/transformers-nodejs-lite": "^2.17.1-lite.4",
+		"@echogarden/wave-codec": "^0.3.3",
 		"@mozilla/readability": "^0.6.0",
 		"chalk": "^5.4.1",
-		"cldr-segmentation": "^2.2.1",
 		"command-exists": "^1.2.9",
-		"compromise": "^14.14.4",
 		"fs-extra": "^11.3.0",
 		"gaxios": "^6.7.1",
 		"graceful-fs": "^4.2.11",
@@ -84,14 +84,16 @@
 		"kuromoji": "^0.1.2",
 		"microsoft-cognitiveservices-speech-sdk": "^1.43.1",
 		"msgpack-lite": "^0.1.26",
-		"onnxruntime-node": "^1.21.1",
-		"openai": "^4.97.0",
+		"onnxruntime-node": "^1.22.0",
+		"openai": "^4.98.0",
+		"regexp-composer": "^0.3.0",
 		"sam-js": "^0.3.1",
 		"strip-ansi": "^7.1.0",
 		"tar": "^7.4.3",
 		"tiktoken": "^1.0.21",
 		"tinyld": "^1.3.4",
 		"wasm-feature-detect": "^1.8.0",
+		"wasm-heap-manager": "^0.2.2",
 		"ws": "^8.18.2",
 		"wtf_wikipedia": "^10.4.0"
 	},
@@ -112,7 +114,7 @@
 		"@types/graceful-fs": "^4.1.9",
 		"@types/jsdom": "^21.1.7",
 		"@types/msgpack-lite": "^0.1.11",
-		"@types/node": "^22.15.3",
+		"@types/node": "^22.15.18",
 		"@types/tar": "^6.1.13",
 		"@types/ws": "^8.18.1",
 		"ts-json-schema-generator": "^2.4.0",

package/src/alignment/SemanticTextAlignment.ts

@@ -3,7 +3,7 @@ import { Logger } from '../utilities/Logger.js'
 import { loadPackage } from '../utilities/PackageManager.js'
 import { alignDTWWindowed } from './DTWSequenceAlignmentWindowed.js'
 import { cosineDistance } from '../math/VectorMath.js'
-import { isPunctuation, isWord, splitToSentences, splitToWords } from '../nlp/Segmentation.js'
+import { includesPunctuation, isWord, parseText } from '../nlp/Segmentation.js'
 import { Timeline, extractEntries } from '../utilities/Timeline.js'
 
 export async function alignTimelineToTextSemantically(timeline: Timeline, text: string, textLangCode: string) {
@@ -26,13 +26,12 @@ export async function alignTimelineToTextSemantically(timeline: Timeline, text:
 
 	const timelineWordEntriesFiltered = timelineWordEntryGroups.flat()
 
-	const textSentences = splitToSentences(text, textLangCode)
+	const segmentedText = await parseText(text, textLangCode)
 
 	const textWordGroups: string[][] = []
 
-	for (const sentenceText of textSentences) {
-		let wordGroup = await splitToWords(sentenceText, textLangCode)
-		wordGroup = wordGroup.filter(word => isWord(word))
+	for (const sentenceEntry of segmentedText.sentences) {
+		const wordGroup = sentenceEntry.words.nonPunctuationWords
 
 		textWordGroups.push(wordGroup)
 	}
@@ -197,8 +196,8 @@ export async function alignWordsToWordsSemantically(wordsGroups1: string[][], wo
 
 	// Align
 	function costFunction(a: TokenEmbeddingData, b: TokenEmbeddingData) {
-		const aIsPunctuation = isPunctuation(a.text)
-		const bIsPunctuation = isPunctuation(b.text)
+		const aIsPunctuation = includesPunctuation(a.text)
+		const bIsPunctuation = includesPunctuation(b.text)
 
 		if (aIsPunctuation === bIsPunctuation) {
 			return cosineDistance(a.embeddingVector, b.embeddingVector)

package/src/api/Alignment.ts

@@ -5,7 +5,7 @@ import { AudioSourceParam, RawAudio, ensureRawAudio, getRawAudioDuration, normal
 import { Logger } from '../utilities/Logger.js'
 
 import * as API from './API.js'
-import { Timeline, addTimeOffsetToTimeline, addWordTextOffsetsToTimeline, wordTimelineToSegmentSentenceTimeline } from '../utilities/Timeline.js'
+import { Timeline, addTimeOffsetToTimeline, addWordTextOffsetsToTimelineInPlace, wordTimelineToSegmentSentenceTimeline } from '../utilities/Timeline.js'
 import { formatLanguageCodeWithName, getDefaultDialectForLanguageCodeIfPossible, getShortLanguageCode, parseLangIdentifier } from '../utilities/Locale.js'
 import { type WhisperAlignmentOptions } from '../recognition/WhisperSTT.js'
 import chalk from 'chalk'
@@ -329,7 +329,7 @@ export async function align(input: AudioSourceParam, transcript: string, options
 	}
 
 	// Add text offsets
-	addWordTextOffsetsToTimeline(mappedTimeline, transcript)
+	addWordTextOffsetsToTimelineInPlace(mappedTimeline, transcript)
 
 	// Make segment timeline
 	const { segmentTimeline } = await wordTimelineToSegmentSentenceTimeline(mappedTimeline, transcript, language, options.plainText?.paragraphBreaks, options.plainText?.whitespace)

package/src/api/Recognition.ts

@@ -5,7 +5,7 @@ import { AudioSourceParam, RawAudio, ensureRawAudio, normalizeAudioLevelInPlace,
 import { Logger } from '../utilities/Logger.js'
 
 import * as API from './API.js'
-import { Timeline, addWordTextOffsetsToTimeline, wordTimelineToSegmentSentenceTimeline } from '../utilities/Timeline.js'
+import { Timeline, addWordTextOffsetsToTimelineInPlace, wordTimelineToSegmentSentenceTimeline } from '../utilities/Timeline.js'
 import { formatLanguageCodeWithName, parseLangIdentifier } from '../utilities/Locale.js'
 import { loadPackage } from '../utilities/PackageManager.js'
 import chalk from 'chalk'
@@ -312,7 +312,7 @@ export async function recognize(input: AudioSourceParam, options: RecognitionOpt
 	}
 
 	// Add text offsets
-	addWordTextOffsetsToTimeline(timeline, transcript)
+	addWordTextOffsetsToTimelineInPlace(timeline, transcript)
 
 	// Make segment timeline
 	const { segmentTimeline } = await wordTimelineToSegmentSentenceTimeline(timeline, transcript, languageCode, 'single', 'preserve')

package/src/api/SpeechTranslation.ts

@@ -4,7 +4,7 @@ import { logToStderr } from '../utilities/Utilities.js'
 import { AudioSourceParam, RawAudio, ensureRawAudio, normalizeAudioLevelInPlace, trimAudioEnd } from '../audio/AudioUtilities.js'
 import { Logger } from '../utilities/Logger.js'
 
-import { Timeline, addWordTextOffsetsToTimeline, wordTimelineToSegmentSentenceTimeline } from '../utilities/Timeline.js'
+import { Timeline, addWordTextOffsetsToTimelineInPlace, wordTimelineToSegmentSentenceTimeline } from '../utilities/Timeline.js'
 import { type WhisperOptions } from '../recognition/WhisperSTT.js'
 import { formatLanguageCodeWithName, getShortLanguageCode, normalizeIdentifierToLanguageCode, parseLangIdentifier } from '../utilities/Locale.js'
 import { EngineMetadata } from './Common.js'
@@ -201,7 +201,7 @@ export async function translateSpeech(input: AudioSourceParam, options: SpeechTr
 	}
 
 	if (wordTimeline) {
-		addWordTextOffsetsToTimeline(wordTimeline, transcript)
+		addWordTextOffsetsToTimelineInPlace(wordTimeline, transcript)
 	}
 
 	if (!segmentTimeline) {

package/src/api/Synthesis.ts

@@ -6,7 +6,7 @@ import { clip, sha256AsHex, stringifyAndFormatJson, logToStderr, yieldToEventLoo
 import { RawAudio, concatAudioSegments, downmixToMono, encodeRawAudioToWave, getSamplePeakDecibels, getEmptyRawAudio, getRawAudioDuration, trimAudioEnd, trimAudioStart, attenuateIfClippingInPlace, normalizeAudioLevelInPlace } from '../audio/AudioUtilities.js'
 import { Logger } from '../utilities/Logger.js'
 
-import { isWordOrSymbolWord, splitToParagraphs, splitToSentences } from '../nlp/Segmentation.js'
+import { isWordOrSymbolWord, parseText, splitToParagraphs } from '../nlp/Segmentation.js'
 import { type RubberbandOptions } from '../dsp/Rubberband.js'
 import { loadLexiconsForLanguage } from '../nlp/Lexicon.js'
 
@@ -116,9 +116,9 @@ async function synthesizeSegments(segments: string[], options: SynthesisOptions,
 	let timeOffset = 0
 
 	for (let segmentIndex = 0; segmentIndex < segments.length; segmentIndex++) {
-		const segmentText = segments[segmentIndex].trim()
+		const segmentText = segments[segmentIndex]
 
-		logger.log(`\n${chalk.magentaBright(`Synthesizing segment ${segmentIndex + 1}/${segments.length}`)}: '${segmentText}'`)
+		logger.log(`\n${chalk.magentaBright(`Synthesizing segment ${segmentIndex + 1}/${segments.length}`)}: '${segmentText.trim()}'`)
 
 		const segmentStartTime = timeOffset
 
@@ -132,9 +132,9 @@ async function synthesizeSegments(segments: string[], options: SynthesisOptions,
 
 		let sentences: string[]
 
-		if ((options.splitToSentences || options.engine == 'vits') && !options.ssml) {
-			sentences = splitToSentences(segmentText, options.language!)
-			sentences = sentences.filter(sentence => sentence.trim() != '')
+		if ((options.splitToSentences || options.engine === 'vits' || options.engine === 'kokoro') && !options.ssml) {
+			const parsedText = await parseText(segmentText, options.language!)
+			sentences = parsedText.sentences.map(sentenceEntry => sentenceEntry.text)
 
 			if (sentences.length == 0) {
 				sentences = ['']
@@ -156,7 +156,7 @@ async function synthesizeSegments(segments: string[], options: SynthesisOptions,
 
 			const sentenceText = sentences[sentenceIndex].trim()
 
-			logger.log(`\n${chalk.magentaBright(`Synthesizing sentence ${sentenceIndex + 1}/${sentences.length}`)}: "${sentenceText}"`)
+			logger.log(`\n${chalk.magentaBright(`Synthesizing sentence ${sentenceIndex + 1}/${sentences.length}`)}: "${sentenceText.trim()}"`)
 
 			const sentenceStartTime = timeOffset
 

package/src/api/TimelineTranslationAlignment.ts

@@ -4,7 +4,7 @@ import { SubtitlesConfig } from '../subtitles/Subtitles.js'
 import { formatLanguageCodeWithName, getShortLanguageCode, parseLangIdentifier } from '../utilities/Locale.js'
 import { Logger } from '../utilities/Logger.js'
 import { extendDeep } from '../utilities/ObjectUtilities.js'
-import { Timeline, addWordTextOffsetsToTimeline, wordTimelineToSegmentSentenceTimeline } from '../utilities/Timeline.js'
+import { Timeline, addWordTextOffsetsToTimelineInPlace, wordTimelineToSegmentSentenceTimeline } from '../utilities/Timeline.js'
 import * as API from './API.js'
 
 export async function alignTimelineTranslation(inputTimeline: Timeline, translatedTranscript: string, options: TimelineTranslationAlignmentOptions): Promise<TimelineTranslationAlignmentResult> {
@@ -95,7 +95,7 @@ export async function alignTimelineTranslation(inputTimeline: Timeline, translat
 
 	logger.start(`Postprocess timeline`)
 
-	addWordTextOffsetsToTimeline(mappedWordTimeline, translatedTranscript)
+	addWordTextOffsetsToTimelineInPlace(mappedWordTimeline, translatedTranscript)
 
 	const { segmentTimeline: mappedTimeline } = await wordTimelineToSegmentSentenceTimeline(mappedWordTimeline, translatedTranscript, targetLanguage)
 

package/src/api/TranslationAlignment.ts

@@ -5,7 +5,7 @@ import { AudioSourceParam, RawAudio, ensureRawAudio, normalizeAudioLevelInPlace,
 import { Logger } from '../utilities/Logger.js'
 
 import * as API from './API.js'
-import { Timeline, addWordTextOffsetsToTimeline, wordTimelineToSegmentSentenceTimeline } from '../utilities/Timeline.js'
+import { Timeline, addWordTextOffsetsToTimelineInPlace, wordTimelineToSegmentSentenceTimeline } from '../utilities/Timeline.js'
 import { formatLanguageCodeWithName, getShortLanguageCode, normalizeIdentifierToLanguageCode, parseLangIdentifier } from '../utilities/Locale.js'
 import { type WhisperAlignmentOptions } from '../recognition/WhisperSTT.js'
 import chalk from 'chalk'
@@ -124,7 +124,7 @@ export async function alignTranslation(input: AudioSourceParam, translatedTransc
 	}
 
 	// Add text offsets
-	addWordTextOffsetsToTimeline(mappedTimeline, translatedTranscript)
+	addWordTextOffsetsToTimelineInPlace(mappedTimeline, translatedTranscript)
 
 	// Make segment timeline
 	const { segmentTimeline } = await wordTimelineToSegmentSentenceTimeline(mappedTimeline, translatedTranscript, sourceLanguage, options.plainText?.paragraphBreaks, options.plainText?.whitespace)

package/src/audio/AudioPlayer.ts

@@ -9,7 +9,7 @@ import { Timer } from '../utilities/Timer.js'
 import { clip, getRandomHexString, waitTimeout, writeToStderr } from '../utilities/Utilities.js'
 import { encodeToAudioBuffer, float32ToInt16Pcm, interleaveChannels } from './AudioBufferConversion.js'
 import { OpenPromise } from '../utilities/OpenPromise.js'
-import { Timeline, addWordTextOffsetsToTimeline } from '../utilities/Timeline.js'
+import { Timeline, addWordTextOffsetsToTimelineInPlace } from '../utilities/Timeline.js'
 import { readAndParseJsonFile, readFileAsUtf8, remove, writeFile } from '../utilities/FileSystem.js'
 import { tryResolvingSoxPath } from './SoxPath.js'
 import { SignalChannel } from '../utilities/SignalChannel.js'
@@ -38,7 +38,7 @@ export async function playAudioWithWordTimeline(rawAudio: RawAudio, wordTimeline
 
 	wordTimeline = deepClone(wordTimeline)
 
-	addWordTextOffsetsToTimeline(wordTimeline, transcript)
+	addWordTextOffsetsToTimelineInPlace(wordTimeline, transcript)
 
 	let timelineEntryIndex = 0
 	let transcriptOffset = 0

package/src/cli/CLI.ts

@@ -10,10 +10,10 @@ import { SubtitlesConfig, subtitlesToText, timelineToSubtitles } from '../subtit
 import { Logger, resetActiveLogger } from '../utilities/Logger.js'
 import { isMainThread, parentPort } from 'node:worker_threads'
 import { encodeFromChannels, getDefaultFFMpegOptionsForSpeech } from '../codecs/FFMpegTranscoder.js'
-import { splitToParagraphs, splitToWords, wordCharacterPattern } from '../nlp/Segmentation.js'
+import { splitToParagraphs, splitToWords } from '../nlp/Segmentation.js'
 import { playAudioSamplesWithKeyboardControls, playAudioWithWordTimeline } from '../audio/AudioPlayer.js'
 import { extendDeep } from '../utilities/ObjectUtilities.js'
-import { Timeline, TimelineEntry, addTimeOffsetToTimeline, addWordTextOffsetsToTimeline, roundTimelineProperties } from '../utilities/Timeline.js'
+import { Timeline, TimelineEntry, addTimeOffsetToTimeline, addWordTextOffsetsToTimelineInPlace, roundTimelineProperties } from '../utilities/Timeline.js'
 import { ensureDir, existsSync, readAndParseJsonFile, readdir, readFileAsUtf8, writeFileSafe } from '../utilities/FileSystem.js'
 import { formatLanguageCodeWithName, getShortLanguageCode } from '../utilities/Locale.js'
 import { APIOptions } from '../api/APIOptions.js'
@@ -509,7 +509,7 @@ export async function speak(operationData: CLIOperationData) {
 	const { audio: synthesizedAudio, timeline } = await API.synthesize(textSegments, options, onSegment, undefined)
 
 	if (plainText) {
-		addWordTextOffsetsToTimeline(timeline, plainText)
+		addWordTextOffsetsToTimelineInPlace(timeline, plainText)
 	}
 
 	if (outputFilenames.length > 0) {
@@ -1884,7 +1884,7 @@ async function checkOutputFilenames(outputFilenames: string[], acceptMediaOutput
 async function writeOutputFilesForSegment(outputFilenames: string[], index: number, total: number, audio: RawAudio, timeline: Timeline, text: string, language: string, allowOverwrite: boolean) {
 	const digitCount = Math.max((total + 1).toString().length, 2)
 
-	const segmentWords = (await splitToWords(text, language)).filter(text => wordCharacterPattern.test(text))
+	const segmentWords = (await splitToWords(text, language)).nonPunctuationWords
 
 	const segmentJoinedWords = segmentWords.join(' ').trim()
 

package/src/denoising/RNNoise.ts

@@ -1,6 +1,6 @@
 import { float32ToInt16Pcm } from '../audio/AudioBufferConversion.js'
 import { concatFloat32Arrays } from '../utilities/Utilities.js'
-import { WasmMemoryManager } from '../utilities/WasmMemoryManager.js'
+import { wrapEmscriptenModuleHeap } from 'wasm-heap-manager'
 import { Logger } from '../utilities/Logger.js'
 import { RawAudio, cloneRawAudio } from '../audio/AudioUtilities.js'
 
@@ -24,15 +24,15 @@ export async function denoiseAudio(rawAudio: RawAudio) {
 	const m = await getRnnoiseInstance()
 
 	logger.start('Process with RNNoise')
-	const wasmMemory = new WasmMemoryManager(m)
+	const wasmHeap = wrapEmscriptenModuleHeap(m)
 
 	const stateSize = m._rnnoise_get_size()
 	const frameSize = m._rnnoise_get_frame_size()
 
 	const denoiseState = m._rnnoise_create(0)
 
-	const inputRef = wasmMemory.allocFloat32Array(frameSize)
-	const outputRef = wasmMemory.allocFloat32Array(frameSize)
+	const inputRef = wasmHeap.allocFloat32Array(frameSize)
+	const outputRef = wasmHeap.allocFloat32Array(frameSize)
 
 	const floatSamples = rawAudio.audioChannels[0]
 	const int16Samples = float32ToInt16Pcm(floatSamples)
@@ -70,7 +70,7 @@ export async function denoiseAudio(rawAudio: RawAudio) {
 	outputNewFrame(outputRef.view.slice(), lastFrameVadProbability)
 
 	m._rnnoise_destroy(denoiseState)
-	wasmMemory.freeAll()
+	wasmHeap.freeAll()
 
 	const int16DenoisedSamplesAsFloats = concatFloat32Arrays(processedFrames)
 

package/src/dsp/FFT.ts

@@ -1,6 +1,6 @@
 import { ComplexNumber } from '../math/VectorMath.js'
 import { concatFloat32Arrays, isWasmSimdSupported } from '../utilities/Utilities.js'
-import { WasmMemoryManager } from '../utilities/WasmMemoryManager.js'
+import { createWasmHeapManager } from 'wasm-heap-manager'
 
 // Compute short-term Fourier transform (real-valued)
 export async function stftr(samples: Float32Array, fftOrder: number, windowSize: number, hopSize: number, windowType: WindowType) {
@@ -15,7 +15,7 @@ export async function stftr(samples: Float32Array, fftOrder: number, windowSize:
 
 // Incrementally generate short-term Fourier transform frames (real-valued)
 export async function* stftrGenerator(samples: Float32Array, fftOrder: number, windowSize: number, hopSize: number, windowType: WindowType) {
-	if (fftOrder % 2 != 0 || windowSize % 2 != 0) {
+	if (fftOrder % 2 !== 0 || windowSize % 2 !== 0) {
 		throw new Error('FFT order and window size must be multiples of 2')
 	}
 
@@ -35,17 +35,19 @@ export async function* stftrGenerator(samples: Float32Array, fftOrder: number, w
 	const windowWeights = getWindowWeights(windowType, windowSize)
 
 	const m = await getPFFFTInstance(await isPffftSimdSupportedForFFTOrder(fftOrder))
-	const wasmMemory = new WasmMemoryManager(m, {
-		wasmAlloc: m._pffft_aligned_malloc,
-		wasmFree: m._pffft_aligned_free
-	})
+
+	const wasmHeap = createWasmHeapManager(
+		() => m.HEAPU8.buffer,
+		m._pffft_aligned_malloc,
+		m._pffft_aligned_free,
+	)
 
 	const statePtr = m._pffft_new_setup(fftOrder, 0)
 
 	const sampleCount = samples.length
-	const frameBufferRef = wasmMemory.allocFloat32Array(fftOrder)
-	const binsBufferRef = wasmMemory.allocFloat32Array(fftOrder * 2)
-	const workBufferRef = wasmMemory.allocFloat32Array(fftOrder * 2)
+	const frameBufferRef = wasmHeap.allocFloat32Array(fftOrder)
+	const binsBufferRef = wasmHeap.allocFloat32Array(fftOrder * 2)
+	const workBufferRef = wasmHeap.allocFloat32Array(fftOrder * 2)
 
 	for (let offset = 0; offset < sampleCount; offset += hopSize) {
 		const windowSamples = samples.subarray(offset, offset + windowSize)
@@ -68,12 +70,12 @@ export async function* stftrGenerator(samples: Float32Array, fftOrder: number, w
 
 	m._pffft_destroy_setup(statePtr)
 
-	wasmMemory.freeAll()
+	wasmHeap.freeAll()
 }
 
 // Compute short-term inverse Fourier transform (real-valued)
 export async function stiftr(binsForFrames: Float32Array[], fftOrder: number, windowSize: number, hopSize: number, windowType: WindowType, expectedOutputLength?: number) {
-	if (fftOrder % 2 != 0 || windowSize % 2 != 0) {
+	if (fftOrder % 2 !== 0 || windowSize % 2 !== 0) {
 		throw new Error('FFT order and window size must multiples of 2')
 	}
 
@@ -101,16 +103,17 @@ export async function stiftr(binsForFrames: Float32Array[], fftOrder: number, wi
 
 	const m = await getPFFFTInstance(await isPffftSimdSupportedForFFTOrder(fftOrder))
 
-	const wasmMemory = new WasmMemoryManager(m, {
-		wasmAlloc: m._pffft_aligned_malloc,
-		wasmFree: m._pffft_aligned_free
-	})
+	const wasmHeap = createWasmHeapManager(
+		() => m.HEAPU8.buffer,
+		m._pffft_aligned_malloc,
+		m._pffft_aligned_free,
+	)
 
 	const statePtr = m._pffft_new_setup(fftOrder, 0)
 
-	const frameBufferRef = wasmMemory.allocFloat32Array(fftOrder)
-	const binsRef = wasmMemory.allocFloat32Array(fftOrder * 2)
-	const workBufferRef = wasmMemory.allocFloat32Array(fftOrder * 2)
+	const frameBufferRef = wasmHeap.allocFloat32Array(fftOrder)
+	const binsRef = wasmHeap.allocFloat32Array(fftOrder * 2)
+	const workBufferRef = wasmHeap.allocFloat32Array(fftOrder * 2)
 
 	const sumOfSquaredWeightsForSample = new Float32Array(outSampleCount)
 
@@ -140,7 +143,7 @@ export async function stiftr(binsForFrames: Float32Array[], fftOrder: number, wi
 	}
 
 	m._pffft_destroy_setup(statePtr)
-	wasmMemory.freeAll()
+	wasmHeap.freeAll()
 
 	// Divide each output sample by the sum of squared weights
 	for (let i = 0; i < outSamples.length; i++) {

package/src/dsp/Rubberband.ts

@@ -1,7 +1,7 @@
 import { RawAudio } from '../audio/AudioUtilities.js'
 import { extendDeep } from '../utilities/ObjectUtilities.js'
 import { concatFloat32Arrays } from '../utilities/Utilities.js'
-import { Float32ArrayRef, WasmMemoryManager } from '../utilities/WasmMemoryManager.js'
+import { Float32ArrayRef, wrapEmscriptenModuleHeap } from 'wasm-heap-manager'
 
 let rubberbandInstance: any
 
@@ -14,7 +14,7 @@ export async function stretchTimePitch(rawAudio: RawAudio, speed: number, pitchS
 	const sampleRate = rawAudio.sampleRate
 
 	const m = await getRubberbandInstance()
-	const wasmMemory = new WasmMemoryManager(m)
+	const wasmHeap = wrapEmscriptenModuleHeap(m)
 
 	const optionFlags = rubberBandOptionsToFlags(options)
 
@@ -25,11 +25,11 @@ export async function stretchTimePitch(rawAudio: RawAudio, speed: number, pitchS
 	const samplesRequired = m._rubberband_get_samples_required(statePtr)
 	const bufferSize = Math.min(samplesRequired, sampleCount)
 
-	const bufferChannelPtrsRef = wasmMemory.allocUint32Array(bufferSize)
+	const bufferChannelPtrsRef = wasmHeap.allocUint32Array(bufferSize)
 	const bufferChannelRefs: Float32ArrayRef[] = []
 
 	for (let i = 0; i < channelCount; i++) {
-		const bufferChannelRef = wasmMemory.allocFloat32Array(bufferSize)
+		const bufferChannelRef = wasmHeap.allocFloat32Array(bufferSize)
 		bufferChannelPtrsRef.view[i] = bufferChannelRef.address
 
 		bufferChannelRefs.push(bufferChannelRef)
@@ -102,7 +102,7 @@ export async function stretchTimePitch(rawAudio: RawAudio, speed: number, pitchS
 	}
 
 	m._rubberband_delete(statePtr)
-	wasmMemory.freeAll()
+	wasmHeap.freeAll()
 
 	const outputAudioChannels = outputAudioChannelChunks.map(chunks => concatFloat32Arrays(chunks))
 	const outputRawAudio: RawAudio = { audioChannels: outputAudioChannels, sampleRate }

package/src/dsp/Sonic.ts

@@ -1,5 +1,5 @@
 import { RawAudio } from '../audio/AudioUtilities.js'
-import { WasmMemoryManager } from '../utilities/WasmMemoryManager.js'
+import { wrapEmscriptenModuleHeap } from 'wasm-heap-manager'
 
 let sonicInstance: any
 
@@ -10,13 +10,13 @@ export async function stretchTimePitch(rawAudio: RawAudio, speed: number, pitchS
 	const inputSampleCount = rawAudio.audioChannels[0].length
 
 	const m = await getSonicInstance()
-	const wasmMemory = new WasmMemoryManager(m)
+	const wasmHeap = wrapEmscriptenModuleHeap(m)
 
 	const streamPtr = m._sonicCreateStream(sampleRate, channelCount)
 	m._sonicSetSpeed(streamPtr, speed)
 	m._sonicSetPitch(streamPtr, pitchScale)
 
-	const inputSamplesRef = wasmMemory.allocFloat32Array(inputSampleCount)
+	const inputSamplesRef = wasmHeap.allocFloat32Array(inputSampleCount)
 	inputSamplesRef.view.set(inputSamples)
 
 	const writeSuccess = m._sonicWriteFloatToStream(streamPtr, inputSamplesRef.address, inputSampleCount)
@@ -33,9 +33,9 @@ export async function stretchTimePitch(rawAudio: RawAudio, speed: number, pitchS
 
 	const samplesAvailable = m._sonicSamplesAvailable(streamPtr)
 
-	const outputSamplesRef = wasmMemory.allocFloat32Array(samplesAvailable)
+	const outputSamplesRef = wasmHeap.allocFloat32Array(samplesAvailable)
 
-	const samplesRead = m._sonicReadFloatFromStream(streamPtr, outputSamplesRef.address, outputSamplesRef.length)
+	const samplesRead = m._sonicReadFloatFromStream(streamPtr, outputSamplesRef.address, outputSamplesRef.allocatedByteCount)
 
 	const outputSamples = outputSamplesRef.view.slice(0, samplesRead)
 
@@ -43,7 +43,7 @@ export async function stretchTimePitch(rawAudio: RawAudio, speed: number, pitchS
 
 	m._sonicDestroyStream(streamPtr)
 
-	wasmMemory.freeAll()
+	wasmHeap.freeAll()
 
 	return resultAudio
 }

package/src/dsp/SpeexResampler.ts

@@ -1,6 +1,6 @@
 import { RawAudio, cloneRawAudio } from '../audio/AudioUtilities.js'
 import { concatFloat32Arrays, isWasmSimdSupported } from '../utilities/Utilities.js'
-import { WasmMemoryManager } from '../utilities/WasmMemoryManager.js'
+import { wrapEmscriptenModuleHeap } from 'wasm-heap-manager'
 
 let speexResamplerInstance: any
 
@@ -23,17 +23,17 @@ export async function resampleAudioSpeex(rawAudio: RawAudio, outSampleRate: numb
 	}
 
 	const m = await getSpeexResamplerInstance()
-	const wasmMemory = new WasmMemoryManager(m)
+	const wasmHeap = wrapEmscriptenModuleHeap(m)
 
 	function speexResultCodeToString(resultCode: number) {
 		const errorStrPtr = m._speex_resampler_strerror(resultCode)
-		const errorStrRef = wasmMemory.wrapUint8Array(errorStrPtr, 1024)
-		const message = errorStrRef.readAsNullTerminatedUtf8String()
+		const messageRef = wasmHeap.wrapNullTerminatedUtf8String(errorStrPtr)
+		const message = messageRef.value
 
 		return message
 	}
 
-	const initErrRef = wasmMemory.allocInt32()
+	const initErrRef = wasmHeap.allocInt32()
 	const resamplerStateAddress = m._speex_resampler_init(channelCount, inSampleRate, outSampleRate, quality, initErrRef.address)
 	let resultCode = initErrRef.value
 
@@ -46,11 +46,11 @@ export async function resampleAudioSpeex(rawAudio: RawAudio, outSampleRate: numb
 
 	const maxChunkSize = 2 ** 20
 
-	const inputChunkSampleCountRef = wasmMemory.allocInt32()
-	const outputChunkSampleCountRef = wasmMemory.allocInt32()
+	const inputChunkSampleCountRef = wasmHeap.allocInt32()
+	const outputChunkSampleCountRef = wasmHeap.allocInt32()
 
-	const inputChunkSamplesRef = wasmMemory.allocFloat32Array(maxChunkSize * 2)
-	const outputChunkSamplesRef = wasmMemory.allocFloat32Array(Math.floor(maxChunkSize * sampleRateRatio) * 2)
+	const inputChunkSamplesRef = wasmHeap.allocFloat32Array(maxChunkSize * 2)
+	const outputChunkSamplesRef = wasmHeap.allocFloat32Array(Math.floor(maxChunkSize * sampleRateRatio) * 2)
 
 	const resampledAudioChunksForChannels: Float32Array[][] = []
 
@@ -65,7 +65,7 @@ export async function resampleAudioSpeex(rawAudio: RawAudio, outSampleRate: numb
 			const inputPaddingSize = isLastChunk ? inputLatency : 0
 			const maxSamplesToRead = Math.min(maxChunkSize, totalSampleCount - readOffset) + inputPaddingSize
 
-			const maxSamplesToWrite = outputChunkSamplesRef.length
+			const maxSamplesToWrite = outputChunkSamplesRef.elementCount
 
 			const inputChunkSamplesForChannel = rawAudio.audioChannels[channelIndex].slice(readOffset, readOffset + maxSamplesToRead)
 
@@ -91,7 +91,7 @@ export async function resampleAudioSpeex(rawAudio: RawAudio, outSampleRate: numb
 	}
 
 	m._speex_resampler_destroy(resamplerStateAddress)
-	wasmMemory.freeAll()
+	wasmHeap.freeAll()
 
 	const resampledAudio: RawAudio = {
 		audioChannels: [],

package/src/nlp/EspeakPhonemizer.ts

@@ -57,7 +57,7 @@ export async function phonemizeText(text: string, voice: string, substitutionMap
 			.replaceAll('«', ', ')
 			.replaceAll('»', ', ')
 
-	const segmentedText = await Segmentation.parse(text, voice)
+	const segmentedText = await Segmentation.parseTextAndConvertToFragmentObjects(text, voice)
 	const preparedPhrases: string[] = []
 	const phraseBreakers: string[] = []