echogarden 1.6.2 → 1.8.0

package/dist/utilities/WindowedList.js

@@ -0,0 +1,36 @@
+export class WindowedList {
+    maxWindowLength;
+    elements = [];
+    startOffset = 0;
+    constructor(maxWindowLength) {
+        this.maxWindowLength = maxWindowLength;
+    }
+    add(value) {
+        if (this.elements.length === this.maxWindowLength) {
+            this.elements.shift();
+            this.startOffset += 1;
+        }
+        this.elements.push(value);
+    }
+    get(index) {
+        if (index < this.startOffset) {
+            throw new Error(`Index is smaller than to window start offset.`);
+        }
+        if (index >= this.endOffset) {
+            throw new Error(`Index is beyond window end offset.`);
+        }
+        return this.elements[index - this.startOffset];
+    }
+    slice(startIndex, endIndex) {
+        const result = [];
+        endIndex = Math.min(endIndex, this.endOffset);
+        for (let i = startIndex; i < endIndex; i++) {
+            result.push(this.get(i));
+        }
+        return result;
+    }
+    get endOffset() {
+        return this.startOffset + this.elements.length;
+    }
+}
+//# sourceMappingURL=WindowedList.js.map

package/dist/utilities/WindowedList.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"WindowedList.js","sourceRoot":"","sources":["../../src/utilities/WindowedList.ts"],"names":[],"mappings":"AAAA,MAAM,OAAO,YAAY;IAII;IAH5B,QAAQ,GAAQ,EAAE,CAAA;IAClB,WAAW,GAAG,CAAC,CAAA;IAEf,YAA4B,eAAuB;QAAvB,oBAAe,GAAf,eAAe,CAAQ;IACnD,CAAC;IAED,GAAG,CAAC,KAAQ;QACX,IAAI,IAAI,CAAC,QAAQ,CAAC,MAAM,KAAK,IAAI,CAAC,eAAe,EAAE,CAAC;YACnD,IAAI,CAAC,QAAQ,CAAC,KAAK,EAAE,CAAA;YAErB,IAAI,CAAC,WAAW,IAAI,CAAC,CAAA;QACtB,CAAC;QAED,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,CAAA;IAC1B,CAAC;IAED,GAAG,CAAC,KAAa;QAChB,IAAI,KAAK,GAAG,IAAI,CAAC,WAAW,EAAE,CAAC;YAC9B,MAAM,IAAI,KAAK,CAAC,+CAA+C,CAAC,CAAA;QACjE,CAAC;QAED,IAAI,KAAK,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;YAC7B,MAAM,IAAI,KAAK,CAAC,oCAAoC,CAAC,CAAA;QACtD,CAAC;QAED,OAAO,IAAI,CAAC,QAAQ,CAAC,KAAK,GAAG,IAAI,CAAC,WAAW,CAAC,CAAA;IAC/C,CAAC;IAED,KAAK,CAAC,UAAkB,EAAE,QAAgB;QACzC,MAAM,MAAM,GAAQ,EAAE,CAAA;QAEtB,QAAQ,GAAG,IAAI,CAAC,GAAG,CAAC,QAAQ,EAAE,IAAI,CAAC,SAAS,CAAC,CAAA;QAE7C,KAAK,IAAI,CAAC,GAAG,UAAU,EAAE,CAAC,GAAG,QAAQ,EAAE,CAAC,EAAE,EAAE,CAAC;YAC5C,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAA;QACzB,CAAC;QAED,OAAO,MAAM,CAAA;IACd,CAAC;IAED,IAAI,SAAS;QACZ,OAAO,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAA;IAC/C,CAAC;CACD"}

package/docs/Engines.md

@@ -81,7 +81,7 @@ These engines' goal is to match (or "align") a given spoken recording with a giv
 
 **Cloud services (unofficial)**:
 
-* Google Translate (`google-translate`): uses the [Google Translate mobile web UI](https://translate.google.com/m) to translate text from and to any one of its supported languages.
+* Google Translate (`google-translate`): uses the [Google Translate mobile web UI](https://translate.google.com/m) to translate text from and to any one of its 243 supported languages.
 
 ## Speech-to-translated-transcript alignment
 

package/docs/Options.md

@@ -47,7 +47,7 @@ Applies to CLI operations: `speak`, `speak-file`, `speak-url`, `speak-wikipedia`
 
 **VITS**:
 * `vits.speakerId`: speaker ID, for VITS models that support multiple speakers. Defaults to `0`
-* `vits.provider`: ONNX execution provider to use. Can be `cpu` or `dml` (https://microsoft.github.io/DirectML/)-based GPU acceleration - Windows only). Using GPU acceleration for VITS may or may not be faster than CPU, depending on your hardware. Defaults to `cpu`
+* `vits.provider`: ONNX execution provider to use. Can be `cpu`, `dml` ([DirectML](https://microsoft.github.io/DirectML/)-based GPU acceleration - Windows only), or `cuda` (Linux only - requires [CUDA Toolkit 12.x](https://developer.nvidia.com/cuda-downloads) and [cuDNN 9.x](https://developer.nvidia.com/cudnn-downloads) to be installed). Using GPU acceleration for VITS may or may not be faster than CPU, depending on your hardware. Defaults to `cpu`
 
 **eSpeak**:
 * `espeak.rate`: speech rate, in eSpeak units. Overrides `speed` when set
@@ -151,8 +151,8 @@ Applies to CLI operation: `transcribe`, API method: `recognize`
 * `whisper.suppressRepetition`: attempt to suppress decoding of repeating token patterns. Defaults to `true`
 * `whisper.repetitionThreshold`: minimal repetition / compressibility score to cause a part not to be auto-prompted to the next part. Defaults to `2.4`
 * `whisper.decodeTimestampTokens`: enable/disable decoding of timestamp tokens. Setting to `false` can reduce the occurrence of hallucinations and token repetition loops, possibly due to the overall reduction in the number of tokens decoded. This has no impact on the accuracy of timestamps, since they are derived independently using cross-attention weights. However, there are cases where this can cause the model to end a part prematurely, especially in singing and less speech-like voice segments, or when there are multiple speakers. Defaults to `true`
-* `whisper.encoderProvider`: identifier for the ONNX execution provider to use with the encoder model. Can be `cpu`, `dml` ([DirectML](https://microsoft.github.io/DirectML/)-based GPU acceleration - Windows only) or `cuda` (Linux only). In general, GPU-based encoding should be significantly faster. Defaults to `cpu`, or `dml` if available
-* `whisper.decoderProvider`: identifier for the ONNX execution provider to use with the decoder model. Can be `cpu`, `dml` (Windows only) or `cuda` (Linux only). Using GPU acceleration for the decoder may be faster than CPU, especially for larger models, but that depends on your particular combination of CPU and GPU. Defaults to `cpu`, and on Windows, `dml` if available for larger models (`small`, `medium`, `large`)
+* `whisper.encoderProvider`: identifier for the ONNX execution provider to use with the encoder model. Can be `cpu`, `dml` ([DirectML](https://microsoft.github.io/DirectML/)-based GPU acceleration - Windows only), or `cuda` (Linux only - requires [CUDA Toolkit 12.x](https://developer.nvidia.com/cuda-downloads) and [cuDNN 9.x](https://developer.nvidia.com/cudnn-downloads) to be installed). In general, GPU-based encoding should be significantly faster. Defaults to `cpu`, or `dml` if available
+* `whisper.decoderProvider`: identifier for the ONNX execution provider to use with the decoder model. Can be `cpu`, `dml` ([DirectML](https://microsoft.github.io/DirectML/)-based GPU acceleration - Windows only), or `cuda` (Linux only - requires [CUDA Toolkit 12.x](https://developer.nvidia.com/cuda-downloads) and [cuDNN 9.x](https://developer.nvidia.com/cudnn-downloads) to be installed). Using GPU acceleration for the decoder may be faster than CPU, especially for larger models, but that depends on your particular combination of CPU and GPU. Defaults to `cpu`, and on Windows, `dml` if available for larger models (`small`, `medium`, `large`)
 * `whisper.seed`: provide a custom random seed for token selection when temperature is greater than 0. Uses a constant seed by default to ensure reproducibility
 
 **Whisper.cpp**:
@@ -174,7 +174,7 @@ Applies to CLI operation: `transcribe`, API method: `recognize`
 
 **Silero**:
 * `silero.modelPath`: path to a Silero model. Note that latest `en`, `de`, `fr` and `uk` models are automatically installed when needed based on the selected language. This should only be used to manually specify a different model, otherwise specify `language` instead
-* `silero.provider`: ONNX execution provider to use. Can be `cpu` or `dml` (Windows only). Defaults to `cpu`, or `dml` if available
+* `silero.provider`: ONNX execution provider to use. Can be `cpu`, `dml` ([DirectML](https://microsoft.github.io/DirectML/)-based GPU acceleration - Windows only), or `cuda` (Linux only - requires [CUDA Toolkit 12.x](https://developer.nvidia.com/cuda-downloads) and [cuDNN 9.x](https://developer.nvidia.com/cudnn-downloads) to be installed). Defaults to `cpu`, or `dml` if available
 
 **Google Cloud**:
 * `googleCloud.apiKey`: Google Cloud API key (required)
@@ -234,11 +234,11 @@ Applies to CLI operation: `align`, API method: `align`
 Applies to the `whisper` engine only. To provide Whisper options for `dtw-ra`, use `recognition.whisper` instead.
 
 * `whisper.model`: Whisper model to use. Defaults to `tiny` or `tiny.en`
-* `whisper.endTokenThreshold`: minimal probability to accept an end-of-text token for a recognized part. The probability is measured via the softmax between the end-of-text token's logit and the second highest logit. You can try to adjust this threshold in cases the model is ending a part with too few, or many tokens decoded. Defaults to `0.9`. On the last audio part, it is always effectively set to `Infinity`, to ensure the remaining transcript tokens are decoded in full
+* `whisper.endTokenThreshold`: minimal probability to accept an end token for a recognized part. The probability is measured via the softmax between the end token's logit and the second highest logit. You can try to adjust this threshold in cases the model is ending a part with too few, or many tokens decoded. Defaults to `0.9`. On the last audio part, it is always effectively set to `Infinity`, to ensure the remaining transcript tokens are decoded in full
+* `whisper.maxTokensPerPart`: maximum number of tokens to decode per part. Should help avoid edge cases where the model never reaches an end token for the part, which otherwise may cause the model to decode too many tokens and eventually crash. Defaults to 250
 * `whisper.encoderProvider`: encoder ONNX provider. See details in recognition section above
 * `whisper.decoderProvider`: decoder ONNX provider. See details in recognition section above
 
-
 ## Speech-to-text translation
 
 Applies to CLI operation: `translate-speech`, API method: `translateSpeech`
@@ -269,10 +269,16 @@ Applies to CLI operation: `translate-speech`, API method: `translateSpeech`
 
 Applies to CLI operation: `translate-text`, API method: `translateText`
 
+**General**:
 * `engine`: only `google-translate` supported
 * `sourceLanguage`: the source language code for the input text. Auto-detected if not set
 * `targetLanguage`: the target language code for the output text. Required
 * `languageDetection`: language detection options. Optional
+* `plainText`: plain text processing options. Optional
+
+**Google Translate**:
+* `googleTranslate.tld`: top-level domain to request from. Defaults to `com`
+* `googleTranslate.maxCharactersPerPart`: maximum number of characters in each part requested from the server. Defaults to 2000
 
 ## Speech-to-translated-transcript alignment
 
@@ -345,7 +351,7 @@ Applies to CLI operation: `detect-speech-langauge`, API method: `detectSpeechLan
 * `whisper.decoderProvider`: decoder ONNX execution provider. See details in recognition section above
 
 **Silero**:
-* `silero.provider`: ONNX execution provider to use. Can be `cpu` or `dml` (Windows only). Using GPU may be faster, but the initialization overhead is larger. **Note**: `dml` provider seems to be unstable at the moment for this model. Defaults to `cpu`
+* `silero.provider`: ONNX execution provider to use. Can be `cpu`, `dml` ([DirectML](https://microsoft.github.io/DirectML/)-based GPU acceleration - Windows only), or `cuda` (Linux only - requires [CUDA Toolkit 12.x](https://developer.nvidia.com/cuda-downloads) and [cuDNN 9.x](https://developer.nvidia.com/cudnn-downloads) to be installed). Using GPU may be faster, but the initialization overhead is larger. **Note**: `dml` provider seems to be unstable at the moment for this model. Defaults to `cpu`
 
 ### Text language detection
 
@@ -370,7 +376,7 @@ Applies to CLI operation: `detect-voice-activity`, API method: `detectVoiceActiv
 
 **Silero**:
 * `silero.frameDuration`: Silero frame duration (ms). Can be `30`, `60` or `90`. Defaults to `90`
-* `silero.provider`: ONNX provider to use. Can be `cpu` or `dml` (Windows only). Using GPU is likely to be slower than CPU due to inference being independently executed on each audio frame. Defaults to `cpu` (recommended)
+* `silero.provider`: ONNX provider to use. Can be `cpu`, `dml` ([DirectML](https://microsoft.github.io/DirectML/)-based GPU acceleration - Windows only), or `cuda` (Linux only - requires [CUDA Toolkit 12.x](https://developer.nvidia.com/cuda-downloads) and [cuDNN 9.x](https://developer.nvidia.com/cudnn-downloads) to be installed). Using GPU is likely to be slower than CPU due to inference being independently executed on each audio frame. Defaults to `cpu` (recommended)
 
 ## Speech denoising
 
@@ -395,8 +401,8 @@ Applies to CLI operation: `isolate`, API method: `isolate`
 
 **MDX-NET**:
 
-* `mdxNet.model`: model to use. Currently available models are `UVR_MDXNET_1_9703`, `UVR_MDXNET_2_9682`, `UVR_MDXNET_3_9662`, `UVR_MDXNET_KARA`. Defaults to `UVR_MDXNET_1_9703`
-* `mdxNet.provider`: ONNX execution provider to use. Can be `cpu` or `dml` ([DirectML](https://microsoft.github.io/DirectML/), Windows only). **Note**: `dml` provider seems to be unstable with MDX-NET models at the moment. Defaults to `cpu`
+* `mdxNet.model`: model to use. Currently available models are `UVR_MDXNET_1_9703`, `UVR_MDXNET_2_9682`, `UVR_MDXNET_3_9662`, `UVR_MDXNET_KARA`, and higher quality models `UVR_MDXNET_Main` and `Kim_Vocal_2`. Defaults to `UVR_MDXNET_1_9703`
+* `mdxNet.provider`: ONNX execution provider to use. Can be `cpu`, `dml` ([DirectML](https://microsoft.github.io/DirectML/)-based GPU acceleration - Windows only), or `cuda` (Linux only - requires [CUDA Toolkit 12.x](https://developer.nvidia.com/cuda-downloads) and [cuDNN 9.x](https://developer.nvidia.com/cudnn-downloads) to be installed). Defaults to `dml` if available (Windows) or `cpu` (other platforms)
 
 # Common options
 

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "echogarden",
-	"version": "1.6.2",
+	"version": "1.8.0",
 	"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",
@@ -46,22 +46,22 @@
 		"LICENSE.md"
 	],
 	"scripts": {
-		"cli": "node --experimental-wasi-unstable-preview1 --no-warnings ./dist/cli/CLIStarter.js",
-		"server": "node --experimental-wasi-unstable-preview1 --no-warnings  ./dist/server/ServerStarter.js",
 		"test": "node --experimental-wasi-unstable-preview1 --no-warnings --trace-uncaught ./dist/tests/Test.js",
-		"generate-options-schema": "npx ts-json-schema-generator --path ./src/api/APIOptions.ts --unstable --type APIOptions --tsconfig ./tsconfig.json --out ./data/schemas/options.json"
+		"generate-options-schema": "npx ts-json-schema-generator --path ./src/api/APIOptions.ts --unstable --type APIOptions --tsconfig ./tsconfig.json --out ./data/schemas/options.json",
+		"make-tarballs": "node --no-warnings ./dist/build-tools/MakeTarballsForInstalledPackages.js"
 	},
 	"bin": {
 		"echogarden": "./dist/cli/CLILauncher.js"
 	},
 	"dependencies": {
-		"@aws-sdk/client-polly": "^3.665.0",
-		"@aws-sdk/client-transcribe-streaming": "^3.665.0",
+		"@aws-sdk/client-polly": "^3.669.0",
+		"@aws-sdk/client-transcribe-streaming": "^3.669.0",
 		"@echogarden/espeak-ng-emscripten": "^0.1.2",
 		"@echogarden/fasttext-wasm": "^0.1.0",
 		"@echogarden/flite-wasi": "^0.1.1",
 		"@echogarden/fvad-wasm": "^0.1.2",
 		"@echogarden/kissfft-wasm": "^0.1.1",
+		"@echogarden/pffft-wasm": "^0.3.0",
 		"@echogarden/rnnoise-wasm": "^0.1.1",
 		"@echogarden/rubberband-wasm": "^0.1.1",
 		"@echogarden/sonic-wasm": "^0.1.1",
@@ -74,14 +74,14 @@
 		"chalk": "^5.3.0",
 		"cldr-segmentation": "^2.2.1",
 		"command-exists": "^1.2.9",
-		"compromise": "^14.14.1",
+		"compromise": "^14.14.2",
 		"fs-extra": "^11.2.0",
 		"gaxios": "^6.7.1",
 		"graceful-fs": "^4.2.11",
 		"html-escaper": "^3.0.3",
 		"html-to-text": "^9.0.5",
 		"import-meta-resolve": "^4.1.0",
-		"jieba-wasm": "^2.1.1",
+		"jieba-wasm": "^2.2.0",
 		"jsdom": "^25.0.1",
 		"json5": "^2.2.3",
 		"kuromoji": "^0.1.2",
@@ -89,7 +89,7 @@
 		"moving-median": "^1.0.0",
 		"msgpack-lite": "^0.1.26",
 		"onnxruntime-node": "^1.19.2",
-		"openai": "^4.67.1",
+		"openai": "^4.67.3",
 		"sam-js": "^0.3.1",
 		"strip-ansi": "^7.1.0",
 		"tar": "^7.4.3",
@@ -116,11 +116,11 @@
 		"@types/graceful-fs": "^4.1.9",
 		"@types/jsdom": "^21.1.7",
 		"@types/msgpack-lite": "^0.1.11",
-		"@types/node": "^22.7.4",
+		"@types/node": "^22.7.5",
 		"@types/recursive-readdir": "^2.2.4",
 		"@types/tar": "^6.1.13",
 		"@types/ws": "^8.5.12",
 		"ts-json-schema-generator": "^2.3.0",
-		"typescript": "^5.6.2"
+		"typescript": "^5.6.3"
 	}
 }

package/src/alignment/SpeechAlignment.ts

@@ -589,7 +589,7 @@ export async function createAlignmentReferenceUsingEspeakForFragments(fragments:
 	{
 		// Split fragments to chunks, process each chunk individually,
 		// and incrementally merge the chunks to the final result.
-		
+
 		const maxCharactersInChunk = 1000
 
 		let timeOffset = 0

package/src/api/Alignment.ts

@@ -1,7 +1,7 @@
 import { extendDeep } from '../utilities/ObjectUtilities.js'
 
 import { logToStderr } from '../utilities/Utilities.js'
-import { AudioSourceParam, RawAudio, ensureRawAudio, getRawAudioDuration, normalizeAudioLevel, trimAudioEnd } from '../audio/AudioUtilities.js'
+import { AudioSourceParam, RawAudio, ensureRawAudio, getRawAudioDuration, normalizeAudioLevelInPlace, trimAudioEnd } from '../audio/AudioUtilities.js'
 import { Logger } from '../utilities/Logger.js'
 
 import * as API from './API.js'
@@ -56,7 +56,7 @@ export async function align(input: AudioSourceParam, transcript: string, options
 
 	logger.start('Normalize and trim audio')
 
-	sourceRawAudio = normalizeAudioLevel(sourceRawAudio)
+	normalizeAudioLevelInPlace(sourceRawAudio)
 	sourceRawAudio.audioChannels[0] = trimAudioEnd(sourceRawAudio.audioChannels[0])
 
 	logger.end()

package/src/api/Denoising.ts

@@ -1,6 +1,6 @@
 import { extendDeep } from '../utilities/ObjectUtilities.js'
 
-import { AudioSourceParam, RawAudio, applyGainDecibels, ensureRawAudio, getSamplePeakDecibels, mixAudio, normalizeAudioLevel } from '../audio/AudioUtilities.js'
+import { AudioSourceParam, RawAudio, applyGainDecibelsInPlace, ensureRawAudio, getSamplePeakDecibels, mixAudio, normalizeAudioLevelInPlace } from '../audio/AudioUtilities.js'
 import { Logger } from '../utilities/Logger.js'
 
 import { logToStderr } from '../utilities/Utilities.js'
@@ -59,13 +59,14 @@ export async function denoise(input: AudioSourceParam, options: DenoisingOptions
 	const dryMixGainDecibels = options.postProcessing!.dryMixGain!
 
 	const preMixPeakDecibels = getSamplePeakDecibels(denoisedAudio.audioChannels)
-	denoisedAudio = mixAudio(denoisedAudio, applyGainDecibels(resampledRawAudio, dryMixGainDecibels))
+	applyGainDecibelsInPlace(resampledRawAudio, dryMixGainDecibels)
+	denoisedAudio = mixAudio(denoisedAudio, resampledRawAudio)
 	const postMixPeakDecibels = getSamplePeakDecibels(denoisedAudio.audioChannels)
 
 	if (shouldNormalize) {
-		denoisedAudio = normalizeAudioLevel(denoisedAudio, targetPeakDecibels, maxGainIncreaseDecibels)
+		normalizeAudioLevelInPlace(denoisedAudio, targetPeakDecibels, maxGainIncreaseDecibels)
 	} else {
-		denoisedAudio = applyGainDecibels(denoisedAudio, preMixPeakDecibels - postMixPeakDecibels)
+		applyGainDecibelsInPlace(denoisedAudio, preMixPeakDecibels - postMixPeakDecibels)
 	}
 
 	logger.end()

package/src/api/Recognition.ts

@@ -1,7 +1,7 @@
 import { extendDeep } from '../utilities/ObjectUtilities.js'
 
 import { logToStderr } from '../utilities/Utilities.js'
-import { AudioSourceParam, RawAudio, ensureRawAudio, normalizeAudioLevel, trimAudioEnd } from '../audio/AudioUtilities.js'
+import { AudioSourceParam, RawAudio, ensureRawAudio, normalizeAudioLevelInPlace, trimAudioEnd } from '../audio/AudioUtilities.js'
 import { Logger } from '../utilities/Logger.js'
 
 import * as API from './API.js'
@@ -59,7 +59,7 @@ export async function recognize(input: AudioSourceParam, options: RecognitionOpt
 
 	logger.start('Normalize and trim audio')
 
-	sourceRawAudio = normalizeAudioLevel(sourceRawAudio)
+	normalizeAudioLevelInPlace(sourceRawAudio)
 	sourceRawAudio.audioChannels[0] = trimAudioEnd(sourceRawAudio.audioChannels[0])
 
 	const engine = options.engine!
@@ -322,7 +322,7 @@ export interface RecognitionResult {
 
 	timeline: Timeline
 	wordTimeline: Timeline
-	
+
 	language: string
 
 	inputRawAudio: RawAudio

package/src/api/SourceSeparation.ts

@@ -1,4 +1,4 @@
-import { AudioSourceParam, RawAudio, ensureRawAudio, subtractAudio } from '../audio/AudioUtilities.js';
+import { AudioSourceParam, RawAudio, attenuateIfClippingInPlace, ensureRawAudio, subtractAudio } from '../audio/AudioUtilities.js';
 import { Logger } from '../utilities/Logger.js';
 import { extendDeep } from '../utilities/ObjectUtilities.js';
 import { loadPackage } from '../utilities/PackageManager.js';
@@ -6,13 +6,13 @@ import { EngineMetadata } from './Common.js';
 import chalk from 'chalk';
 import { readdir } from '../utilities/FileSystem.js';
 import path from 'node:path';
-import { OnnxExecutionProvider } from '../utilities/OnnxUtilities.js';
+import { defaultMDXNetOptions, getProfileForMDXNetModelName, MDXNetOptions } from '../source-separation/MDXNetSourceSeparation.js';
 
 export async function isolate(input: AudioSourceParam, options: SourceSeparationOptions): Promise<SourceSeparationResult> {
 	const logger = new Logger()
 	const startTimestamp = logger.getTimestamp()
 
-	logger.start('Prepare for source separation')
+	await logger.startAsync('Prepare for source separation')
 
 	const inputRawAudio = await ensureRawAudio(input)
 
@@ -27,8 +27,6 @@ export async function isolate(input: AudioSourceParam, options: SourceSeparation
 
 			const mdxNetOptions = options.mdxNet!
 
-			const executionProviders: OnnxExecutionProvider[] = mdxNetOptions.executionProvider ? [mdxNetOptions.executionProvider] : []
-
 			const packageDir = await loadPackage(`mdxnet-${mdxNetOptions.model!}`)
 			const modelFilename = (await readdir(packageDir)).filter(name => name.endsWith('onnx'))[0]
 
@@ -38,16 +36,20 @@ export async function isolate(input: AudioSourceParam, options: SourceSeparation
 
 			const modelPath = path.join(packageDir, modelFilename)
 
+			await logger.startAsync(`Convert audio to 44.1 kHz stereo`)
+
+			const inputRawAudio44100Stereo = await ensureRawAudio(inputRawAudio, 44100, 2)
+
 			logger.end()
 
-			const audioStereo44100 = await ensureRawAudio(inputRawAudio, 44100, 2)
+			const modelProfile = getProfileForMDXNetModelName(mdxNetOptions.model!)
 
-			isolatedRawAudio = await MDXNetSourceSeparation.isolate(audioStereo44100, modelPath, executionProviders)
+			isolatedRawAudio = await MDXNetSourceSeparation.isolate(inputRawAudio44100Stereo, modelPath, modelProfile, mdxNetOptions)
 
 			logger.end()
 
-			logger.start(`Subtract from original waveform to extract background audio`)
-			backgroundRawAudio = subtractAudio(audioStereo44100, isolatedRawAudio)
+			await logger.startAsync(`Subtract from original waveform to extract background audio`)
+			backgroundRawAudio = subtractAudio(inputRawAudio44100Stereo, isolatedRawAudio)
 
 			break
 		}
@@ -57,8 +59,13 @@ export async function isolate(input: AudioSourceParam, options: SourceSeparation
 		}
 	}
 
+	await logger.startAsync(`Postprocess audio`)
+
+	attenuateIfClippingInPlace(isolatedRawAudio)
+	attenuateIfClippingInPlace(backgroundRawAudio)
+
 	logger.end()
-	logger.log('')
+
 	logger.logDuration(`Total source separation time`, startTimestamp, chalk.magentaBright)
 
 	return {
@@ -73,19 +80,13 @@ export type SourceSeparationEngine = 'mdx-net'
 export interface SourceSeparationOptions {
 	engine?: SourceSeparationEngine
 
-	mdxNet?: {
-		model?: string
-		executionProvider?: OnnxExecutionProvider
-	}
+	mdxNet?: MDXNetOptions
 }
 
 export const defaultSourceSeparationOptions: SourceSeparationOptions = {
 	engine: 'mdx-net',
 
-	mdxNet: {
-		model: 'UVR_MDXNET_1_9703',
-		executionProvider: undefined,
-	}
+	mdxNet: defaultMDXNetOptions,
 }
 
 export interface SourceSeparationResult {
@@ -98,7 +99,7 @@ export const sourceSeparationEngines: EngineMetadata[] = [
 	{
 		id: 'mdx-net',
 		name: 'MDX-NET',
-		description: 'Deep learning source separation architecture by KUIELAB (Korea University).',
+		description: 'Deep learning audio source separation architecture by KUIELAB (Korea University).',
 		type: 'local'
 	},
 ]

package/src/api/SpeechLanguageDetection.ts

@@ -1,6 +1,6 @@
 import { deepClone, extendDeep } from '../utilities/ObjectUtilities.js'
 
-import { AudioSourceParam, RawAudio, ensureRawAudio, getRawAudioDuration, normalizeAudioLevel, sliceRawAudioByTime, trimAudioEnd } from '../audio/AudioUtilities.js'
+import { AudioSourceParam, RawAudio, ensureRawAudio, getRawAudioDuration, normalizeAudioLevelInPlace, sliceRawAudioByTime, trimAudioEnd } from '../audio/AudioUtilities.js'
 import { Logger } from '../utilities/Logger.js'
 
 import * as API from './API.js'
@@ -28,7 +28,7 @@ export async function detectSpeechLanguage(input: AudioSourceParam, options: Spe
 
 	logger.start(`Resample audio to 16kHz mono`)
 	let sourceRawAudio = await ensureRawAudio(inputRawAudio, 16000, 1)
-	sourceRawAudio = normalizeAudioLevel(sourceRawAudio)
+	normalizeAudioLevelInPlace(sourceRawAudio)
 	sourceRawAudio.audioChannels[0] = trimAudioEnd(sourceRawAudio.audioChannels[0])
 
 	if (options.crop) {

package/src/api/SpeechTranslation.ts

@@ -1,7 +1,7 @@
 import { extendDeep } from '../utilities/ObjectUtilities.js'
 
 import { logToStderr } from '../utilities/Utilities.js'
-import { AudioSourceParam, RawAudio, ensureRawAudio, normalizeAudioLevel, trimAudioEnd } from '../audio/AudioUtilities.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'
@@ -61,7 +61,7 @@ export async function translateSpeech(input: AudioSourceParam, options: SpeechTr
 
 	logger.start('Normalize and trim audio')
 
-	sourceRawAudio = normalizeAudioLevel(sourceRawAudio)
+	normalizeAudioLevelInPlace(sourceRawAudio)
 	sourceRawAudio.audioChannels[0] = trimAudioEnd(sourceRawAudio.audioChannels[0])
 
 	if (options.sourceLanguage) {

package/src/api/Synthesis.ts

@@ -4,8 +4,8 @@ import { deepClone, extendDeep } from '../utilities/ObjectUtilities.js'
 
 import * as FFMpegTranscoder from '../codecs/FFMpegTranscoder.js'
 
-import { clip, convertHtmlToText, sha256AsHex, simplifyPunctuationCharacters, stringifyAndFormatJson, logToStderr, yieldToEventLoop, runOperationWithRetries } from '../utilities/Utilities.js'
-import { RawAudio, attenuateIfClipping, concatAudioSegments, downmixToMono, encodeRawAudioToWave, getSamplePeakDecibels, getEmptyRawAudio, getRawAudioDuration, normalizeAudioLevel, trimAudioEnd, trimAudioStart } from '../audio/AudioUtilities.js'
+import { clip, sha256AsHex, stringifyAndFormatJson, logToStderr, yieldToEventLoop, runOperationWithRetries } from '../utilities/Utilities.js'
+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'
@@ -25,6 +25,8 @@ import { type EspeakOptions } from '../synthesis/EspeakTTS.js'
 import { type OpenAICloudTTSOptions  } from '../synthesis/OpenAICloudTTS.js'
 import { type ElevenlabsTTSOptions } from '../synthesis/ElevenlabsTTS.js'
 import { OnnxExecutionProvider } from '../utilities/OnnxUtilities.js'
+import { simplifyPunctuationCharacters } from '../nlp/TextNormalizer.js'
+import { convertHtmlToText } from '../utilities/StringUtilities.js'
 
 const log = logToStderr
 
@@ -248,9 +250,9 @@ async function synthesizeSegments(segments: string[], options: SynthesisOptions,
 		resultRawAudio = { audioChannels: joinedAudioBuffers, sampleRate: segmentsRawAudio[0].sampleRate }
 
 		if (options.postProcessing!.normalizeAudio) {
-			resultRawAudio = normalizeAudioLevel(resultRawAudio, options.postProcessing!.targetPeak, options.postProcessing!.maxGainIncrease)
+			normalizeAudioLevelInPlace(resultRawAudio, options.postProcessing!.targetPeak, options.postProcessing!.maxGainIncrease)
 		} else {
-			resultRawAudio = attenuateIfClipping(resultRawAudio)
+			attenuateIfClippingInPlace(resultRawAudio)
 		}
 	} else {
 		resultRawAudio = getEmptyRawAudio(1, 24000)
@@ -864,9 +866,9 @@ async function synthesizeSegment(text: string, options: SynthesisOptions) {
 	synthesizedAudio = downmixToMono(synthesizedAudio)
 
 	if (options.postProcessing!.normalizeAudio) {
-		synthesizedAudio = normalizeAudioLevel(synthesizedAudio, options.postProcessing!.targetPeak!, options.postProcessing!.maxGainIncrease!)
+		normalizeAudioLevelInPlace(synthesizedAudio, options.postProcessing!.targetPeak!, options.postProcessing!.maxGainIncrease!)
 	} else {
-		synthesizedAudio = attenuateIfClipping(synthesizedAudio)
+		attenuateIfClippingInPlace(synthesizedAudio)
 	}
 
 	const preTrimSampleCount = synthesizedAudio.audioChannels[0].length

package/src/api/TextTranslation.ts

@@ -3,6 +3,7 @@ import { formatLanguageCodeWithName, normalizeIdentifierToLanguageCode, parseLan
 import { Logger } from '../utilities/Logger.js'
 import { extendDeep } from '../utilities/ObjectUtilities.js'
 import * as API from './API.js'
+import { defaultGoogleTranslateTextTranslationOptions, type GoogleTranslateTextTranslationOptions } from '../text-translation/GoogleTranslateTextTranslation.js'
 
 export async function translateText(inputText: string, options: TextTranslationOptions): Promise<TextTranslationResult> {
 	const logger = new Logger()
@@ -43,7 +44,7 @@ export async function translateText(inputText: string, options: TextTranslationO
 
 			logger.end()
 
-			logger.logTitledMessage(`Warning`, `The nllb engine is currently an early prototype implementation and doesn't work correctly.`, chalk.yellow, 'warning')
+			logger.logTitledMessage(`Warning`, `The nllb text translation engine is currently a work-in-progress and doesn't work correctly.`, chalk.yellow, 'warning')
 
 			translationPairs = await NLLBTextTranslation.translateText(inputText, options.sourceLanguage, options.targetLanguage)
 
@@ -63,9 +64,11 @@ export async function translateText(inputText: string, options: TextTranslationO
 		case 'google-translate': {
 			const GoogleTranslateTextTranslation = await import('../text-translation/GoogleTranslateTextTranslation.js')
 
+			const googleTranslateOptions = options.googleTranslate!
+
 			logger.end();
 
-			({ translationPairs, translatedText } = await GoogleTranslateTextTranslation.translateText(inputText, options.sourceLanguage, options.targetLanguage))
+			({ translationPairs, translatedText } = await GoogleTranslateTextTranslation.translateText(inputText, options.sourceLanguage, options.targetLanguage, options.plainText!, googleTranslateOptions))
 
 			break
 		}
@@ -75,7 +78,7 @@ export async function translateText(inputText: string, options: TextTranslationO
 
 			logger.end()
 
-			logger.logTitledMessage(`Warning`, `The deepl engine is currently an early prototype implementation and doesn't work correctly.`, chalk.yellow, 'warning')
+			logger.logTitledMessage(`Warning`, `The deepl text translation engine is currently a work-in-progress and doesn't work correctly.`, chalk.yellow, 'warning')
 
 			translationPairs = await DeepLTextTranslation.translateText(inputText, options.sourceLanguage, options.targetLanguage)
 			translatedText = ''
@@ -113,11 +116,12 @@ export interface TextTranslationOptions {
 
 	languageDetection?: API.TextLanguageDetectionOptions
 
+	plainText?: API.PlainTextOptions
+
 	nllb?: {
 	},
 
-	googleTranslate?: {
-	},
+	googleTranslate?: GoogleTranslateTextTranslationOptions,
 
 	deepl?: {
 	},
@@ -148,12 +152,16 @@ export const defaultTextTranslationOptions: TextTranslationOptions = {
 
 	languageDetection: undefined,
 
-	nllb: {
+	plainText: {
+		paragraphBreaks: 'double',
+		whitespace: 'preserve'
 	},
 
-	googleTranslate: {
+	nllb: {
 	},
 
+	googleTranslate: defaultGoogleTranslateTextTranslationOptions,
+
 	deepl: {
 	},
 }
@@ -162,7 +170,7 @@ export const textTranslationEngines: API.EngineMetadata[] = [
 	{
 		id: 'nllb',
 		name: 'NLLB',
-		description: 'No Language Left Behind (NLLB) is a deep learning machine translation model by Facebook Research (early prototype implementation).',
+		description: 'No Language Left Behind (NLLB) is a deep learning machine translation model by Facebook Research (work-in-progress, do not use).',
 		type: 'local'
 	},
 	{
@@ -174,7 +182,7 @@ export const textTranslationEngines: API.EngineMetadata[] = [
 	{
 		id: 'deepl',
 		name: 'DeepL',
-		description: 'Unoffical text translation API used by the DeepL web interface (early prototype implementation).',
+		description: 'Unoffical text translation API used by the DeepL web interface (work-in-progress, do not use).',
 		type: 'cloud'
 	},
 ]

package/src/api/TranslationAlignment.ts

@@ -1,7 +1,7 @@
 import { extendDeep } from '../utilities/ObjectUtilities.js'
 
 import { logToStderr } from '../utilities/Utilities.js'
-import { AudioSourceParam, RawAudio, ensureRawAudio, normalizeAudioLevel, trimAudioEnd } from '../audio/AudioUtilities.js'
+import { AudioSourceParam, RawAudio, ensureRawAudio, normalizeAudioLevelInPlace, trimAudioEnd } from '../audio/AudioUtilities.js'
 import { Logger } from '../utilities/Logger.js'
 
 import * as API from './API.js'
@@ -53,7 +53,7 @@ export async function alignTranslation(input: AudioSourceParam, translatedTransc
 
 	logger.start('Normalize and trim audio')
 
-	sourceRawAudio = normalizeAudioLevel(sourceRawAudio)
+	normalizeAudioLevelInPlace(sourceRawAudio)
 	sourceRawAudio.audioChannels[0] = trimAudioEnd(sourceRawAudio.audioChannels[0])
 
 	logger.end()