echogarden 2.8.3 → 2.8.4

package/data/lexicons/heteronyms.en.json

@@ -2026,39 +2026,39 @@
 			{
 				"pronunciation": {
 					"espeak": {
-						"en-us": "ɹ ᵻ f j ˈu z",
-						"en-gb-x-rp": "ɹ ᵻ f j ˈuː z"
+						"en-us": "ɹ ˈɛ f j uː s",
+						"en-gb-x-rp": "ɹ ˈɛ f j uː s"
 					}
 				},
 				"pos": [
-					"VB"
+					"NN"
 				],
 
 				"precededBy": [
-					"to", "i", "me", "they", "we", "should", "will", "would", "wouldn't", "won't", "don't", "doesn't", "should", "shouldn't", "could", "couldn't", "must", "mustn't", "shall", "can", "can't", "may", "might", "did", "didn't", "does", "doesn't"
+					"a", "the", "more", "every", "each"
 				],
 
 				"notPrecededBy": [
-					"the"
+					"to", "i", "me", "they", "we", "should", "will", "would", "wouldn't", "won't", "don't", "doesn't", "should", "shouldn't", "could", "couldn't", "must", "mustn't", "shall", "can", "can't", "may", "might", "did", "didn't", "does", "doesn't"
 				],
 
 				"followedBy": [
-					"to"
 				],
 
 				"notFollowedBy": [
+					"to"
 				],
 
-				"example": "I must refuse."
+				"example": "We separate refuse from recyclables."
 			},
 			{
 				"pronunciation": {
 					"espeak": {
-						"en-us": "ɹ ˈɛ f j uː s",
-						"en-gb-x-rp": "ɹ ˈɛ f j uː s"
+						"en-us": "ɹ ᵻ f j ˈu z",
+						"en-gb-x-rp": "ɹ ᵻ f j ˈuː z"
 					}
 				},
-				"example": "We separate refuse from recyclables."
+				"example": "I must refuse."
 			}
 		],
 		"import": [

package/data/lexicons/words.en.json

@@ -1703,6 +1703,46 @@
 					"en-gb-x-rp": "n ə m ˈɛ ŋ k l ə tʃ ɐ"
 				}
 			}
+		},
+		"axes": {
+			"pronunciation": {
+				"espeak": {
+					"en-us": "ˈæ k s i z",
+					"en-gb-x-rp": "ˈæ k s i z"
+				}
+			}
+		},
+		"preloaded": {
+			"pronunciation": {
+				"espeak": {
+					"en-us": "p ɹ i l ˈoʊ d ᵻ d",
+					"en-gb-x-rp": "p ɹ i l ˈoʊ d ᵻ d"
+				}
+			}
+		},
+		"absorbed": {
+			"pronunciation": {
+				"espeak": {
+					"en-us": "ə b z ˈɔːɹ b d",
+					"en-gb-x-rp": "ə b z ˈɔː b d"
+				}
+			}
+		},
+		"disparate": {
+			"pronunciation": {
+				"espeak": {
+					"en-us": "d ˈɪ s p ɹ ə t",
+					"en-gb-x-rp": "d ˈɪ s p ɹ ə t"
+				}
+			}
+		},
+		"qwen": {
+			"pronunciation": {
+				"espeak": {
+					"en-us": "k w ˈɛ n",
+					"en-gb-x-rp": "k w ˈɛ n"
+				}
+			}
 		}
 	}
 }

package/dist/nlp/IPA.d.ts

@@ -1,6 +1,6 @@
-export declare function getPhoneSubstitutionCost1(ipa1: string, ipa2: string): 1 | 0;
-export declare function getPhoneSubstitutionCost2(ipa1: string, ipa2: string): 1 | 0 | 0.5;
-export declare function getPhoneSubstitutionCost3(ipa1: string, ipa2: string): 1 | 0 | 0.5 | 0.75;
+export declare function getPhoneSubstitutionCost1(ipa1: string, ipa2: string): 0 | 1;
+export declare function getPhoneSubstitutionCost2(ipa1: string, ipa2: string): 0 | 1 | 0.5;
+export declare function getPhoneSubstitutionCost3(ipa1: string, ipa2: string): 0 | 1 | 0.5 | 0.75;
 export declare function isKnownSymbol(ipa: string): boolean;
 export declare function isConsonant(ipa: string): boolean;
 export declare function isVowel(ipa: string): boolean;

package/dist/recognition/WhisperSTT.d.ts

@@ -82,7 +82,7 @@ export declare class Whisper {
     languageTokenToLanguageIndex(languageToken: number): void;
     get isEnglishOnlyModel(): boolean;
     get isLargeModel(): boolean;
-    get filterbankCount(): 128 | 80;
+    get filterbankCount(): 80 | 128;
     get filterbanks(): Filterbank[];
     get alignmentHeadIndexes(): number[];
     get defaultTimestampAccuracy(): "medium" | "high";

package/docs/Options.md

@@ -163,7 +163,7 @@ Applies to CLI operation: `transcribe`, API method: `recognize`
 * `whisper.topCandidateCount`: the number of top candidate tokens to consider. Defaults to `5`
 * `whisper.punctuationThreshold`: the minimal probability for a punctuation token, included in the top candidates, to be chosen unconditionally. A lower threshold encourages the model to output more punctuation characters. Defaults to `0.2`
 * `whisper.autoPromptParts`: use previous part's recognized text as the prompt for the next part. Disabling this may help to prevent repetition carrying over between parts, in some cases. Defaults to `true` (**Note**: currently always disabled for `large-v3-turbo` model due to an apparent issue with corrupt output when prompted)
-* `whisper.maxTokensPerPart`: maximum number of tokens to decode for each audio part. Defaults to `250`
+* `whisper.maxTokensPerPart`: maximum number of tokens to decode for each audio part. Defaults to `220` (which is the highest possible)
 * `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`
@@ -260,7 +260,7 @@ Applies to the `whisper` engine only. To provide Whisper options for `dtw-ra`, u
 
 * `whisper.model`: Whisper model to use. Defaults to `tiny` or `tiny.en`
 * `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.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 `220` (highest possible)
 * `whisper.timestampAccuracy`: timestamp accuracy. can be `medium` or `high`. `medium` uses a reduced subset of attention heads for alignment, `high` uses all attention heads and is thus more accurate at the word level, but slower for larger models. Defaults to `high` for the `tiny` and `base` models, and `medium` for the larger models. Note: setting `high` for model `large-v3-turbo` reverts to `medium`, due to poor results with the `high` setting for that model (possibly related to the 16-bit quantization).
 * `whisper.encoderProvider`: encoder ONNX provider. See details in recognition section above
 * `whisper.decoderProvider`: decoder ONNX provider. See details in recognition section above

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "echogarden",
-	"version": "2.8.3",
+	"version": "2.8.4",
 	"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",
@@ -84,7 +84,7 @@
 		"kuromoji": "^0.1.2",
 		"microsoft-cognitiveservices-speech-sdk": "^1.43.1",
 		"msgpack-lite": "^0.1.26",
-		"onnxruntime-node": "^1.22.0",
+		"onnxruntime-node": "^1.21.1",
 		"openai": "^4.98.0",
 		"regexp-composer": "^0.3.0",
 		"sam-js": "^0.3.1",