@memberjunction/ai-groq 6.1.0-edge.0 → 6.1.0-edge.2

package/dist/index.d.ts

@@ -1,2 +1,3 @@
 export * from './models/groq.js';
+export * from './models/groqAudio.js';
 //# sourceMappingURL=index.d.ts.map

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,cAAc,eAAe,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,cAAc,eAAe,CAAC;AAC9B,cAAc,oBAAoB,CAAC"}

package/dist/index.js

@@ -1,2 +1,3 @@
 export * from './models/groq.js';
+export * from './models/groqAudio.js';
 //# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,cAAc,eAAe,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,cAAc,eAAe,CAAC;AAC9B,cAAc,oBAAoB,CAAC"}

package/dist/models/groqAudio.d.ts

@@ -0,0 +1,49 @@
+import { AudioModel, AudioSplitter, BaseAudioGenerator, PronounciationDictionary, SpeechResult, SpeechToTextParams, TextToSpeechParams, VoiceInfo } from '@memberjunction/ai';
+import Groq from 'groq-sdk';
+/**
+ * Groq implementation of {@link BaseAudioGenerator}, covering speech-to-text via Whisper.
+ *
+ * Groq offers no text-to-speech surface, so `CreateSpeech` and the voice/dictionary methods
+ * are not supported — `GetSupportedMethods` reports exactly what works.
+ *
+ * Whisper on Groq does **not** do speaker diarization. There is no parameter for it and no
+ * way to emulate it from the response, so a caller that needs speaker labels needs a
+ * different provider (AssemblyAI, Deepgram) rather than a flag here.
+ *
+ * Audio above Groq's 25MB upload ceiling requires an {@link AudioSplitter}, assigned to
+ * {@link Splitter}. Without one, oversized audio fails with a message naming the option
+ * rather than silently transcribing a truncated prefix.
+ */
+export declare class GroqAudioGenerator extends BaseAudioGenerator {
+    private _client;
+    constructor(apiKey: string);
+    /**
+     * Read only getter for the underlying Groq client instance
+     */
+    get GroqClient(): Groq;
+    /**
+     * Optional splitter used only when the audio exceeds Groq's upload ceiling. See
+     * {@link AudioSplitter} for why this is injected rather than bundled.
+     */
+    Splitter: AudioSplitter | null;
+    /**
+     * Transcribes audio to text using one of Groq's Whisper models.
+     *
+     * Pieces of split audio are transcribed **sequentially**, not in parallel: Groq rate
+     * limits by audio-seconds per minute, so firing an hour of audio at once buys nothing
+     * but 429s, and a partial failure mid-way would leave a transcript with an unmarked
+     * hole in it.
+     */
+    SpeechToText(params: SpeechToTextParams): Promise<SpeechResult>;
+    /**
+     * Resolve the audio bytes from whichever of the two input forms the caller used.
+     */
+    private resolveAudio;
+    private transcribeOne;
+    CreateSpeech(_params: TextToSpeechParams): Promise<SpeechResult>;
+    GetVoices(): Promise<VoiceInfo[]>;
+    GetModels(): Promise<AudioModel[]>;
+    GetPronounciationDictionaries(): Promise<PronounciationDictionary[]>;
+    GetSupportedMethods(): Promise<string[]>;
+}
+//# sourceMappingURL=groqAudio.d.ts.map

package/dist/models/groqAudio.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"groqAudio.d.ts","sourceRoot":"","sources":["../../src/models/groqAudio.ts"],"names":[],"mappings":"AAAA,OAAO,EACH,UAAU,EACV,aAAa,EACb,kBAAkB,EAElB,wBAAwB,EACxB,YAAY,EACZ,kBAAkB,EAClB,kBAAkB,EAClB,SAAS,EACZ,MAAM,oBAAoB,CAAC;AAE5B,OAAO,IAAgB,MAAM,UAAU,CAAC;AAgBxC;;;;;;;;;;;;;GAaG;AACH,qBACa,kBAAmB,SAAQ,kBAAkB;IACtD,OAAO,CAAC,OAAO,CAAO;gBAEV,MAAM,EAAE,MAAM;IAK1B;;OAEG;IACH,IAAW,UAAU,IAAI,IAAI,CAE5B;IAED;;;OAGG;IACI,QAAQ,EAAE,aAAa,GAAG,IAAI,CAAQ;IAE7C;;;;;;;OAOG;IACU,YAAY,CAAC,MAAM,EAAE,kBAAkB,GAAG,OAAO,CAAC,YAAY,CAAC;IAuD5E;;OAEG;IACH,OAAO,CAAC,YAAY;YAUN,aAAa;IAYd,YAAY,CAAC,OAAO,EAAE,kBAAkB,GAAG,OAAO,CAAC,YAAY,CAAC;IAIhE,SAAS,IAAI,OAAO,CAAC,SAAS,EAAE,CAAC;IAKjC,SAAS,IAAI,OAAO,CAAC,UAAU,EAAE,CAAC;IA0BlC,6BAA6B,IAAI,OAAO,CAAC,wBAAwB,EAAE,CAAC;IAIpE,mBAAmB,IAAI,OAAO,CAAC,MAAM,EAAE,CAAC;CAGxD"}

package/dist/models/groqAudio.js

@@ -0,0 +1,177 @@
+var __decorate = (this && this.__decorate) || function (decorators, target, key, desc) {
+    var c = arguments.length, r = c < 3 ? target : desc === null ? desc = Object.getOwnPropertyDescriptor(target, key) : desc, d;
+    if (typeof Reflect === "object" && typeof Reflect.decorate === "function") r = Reflect.decorate(decorators, target, key, desc);
+    else for (var i = decorators.length - 1; i >= 0; i--) if (d = decorators[i]) r = (c < 3 ? d(r) : c > 3 ? d(target, key, r) : d(target, key)) || r;
+    return c > 3 && r && Object.defineProperty(target, key, r), r;
+};
+var __metadata = (this && this.__metadata) || function (k, v) {
+    if (typeof Reflect === "object" && typeof Reflect.metadata === "function") return Reflect.metadata(k, v);
+};
+import { BaseAudioGenerator, ErrorAnalyzer, SpeechResult, } from '@memberjunction/ai';
+import { RegisterClass } from '@memberjunction/global';
+import Groq, { toFile } from 'groq-sdk';
+/**
+ * Groq's documented upload ceiling for the transcription endpoint. The request is rejected
+ * outright above this, so it is a hard boundary rather than a tuning knob.
+ */
+const MAX_UPLOAD_BYTES = 25 * 1024 * 1024;
+/**
+ * Leave headroom under the ceiling: multipart framing and the form fields ride along with
+ * the audio, and a piece sized exactly at the limit fails for the overhead alone.
+ */
+const SPLIT_TARGET_BYTES = 24 * 1024 * 1024;
+const DEFAULT_MODEL = 'whisper-large-v3';
+/**
+ * Groq implementation of {@link BaseAudioGenerator}, covering speech-to-text via Whisper.
+ *
+ * Groq offers no text-to-speech surface, so `CreateSpeech` and the voice/dictionary methods
+ * are not supported — `GetSupportedMethods` reports exactly what works.
+ *
+ * Whisper on Groq does **not** do speaker diarization. There is no parameter for it and no
+ * way to emulate it from the response, so a caller that needs speaker labels needs a
+ * different provider (AssemblyAI, Deepgram) rather than a flag here.
+ *
+ * Audio above Groq's 25MB upload ceiling requires an {@link AudioSplitter}, assigned to
+ * {@link Splitter}. Without one, oversized audio fails with a message naming the option
+ * rather than silently transcribing a truncated prefix.
+ */
+let GroqAudioGenerator = class GroqAudioGenerator extends BaseAudioGenerator {
+    constructor(apiKey) {
+        super(apiKey);
+        /**
+         * Optional splitter used only when the audio exceeds Groq's upload ceiling. See
+         * {@link AudioSplitter} for why this is injected rather than bundled.
+         */
+        this.Splitter = null;
+        this._client = new Groq({ apiKey: apiKey });
+    }
+    /**
+     * Read only getter for the underlying Groq client instance
+     */
+    get GroqClient() {
+        return this._client;
+    }
+    /**
+     * Transcribes audio to text using one of Groq's Whisper models.
+     *
+     * Pieces of split audio are transcribed **sequentially**, not in parallel: Groq rate
+     * limits by audio-seconds per minute, so firing an hour of audio at once buys nothing
+     * but 429s, and a partial failure mid-way would leave a transcript with an unmarked
+     * hole in it.
+     */
+    async SpeechToText(params) {
+        const result = new SpeechResult();
+        try {
+            const audio = this.resolveAudio(params);
+            const model = params.model || DEFAULT_MODEL;
+            let text;
+            if (audio.byteLength <= MAX_UPLOAD_BYTES) {
+                text = await this.transcribeOne(audio, model, params);
+            }
+            else {
+                if (!this.Splitter) {
+                    throw new Error(`Audio is ${(audio.byteLength / (1024 * 1024)).toFixed(1)}MB, above Groq's 25MB ` +
+                        `transcription limit. Assign an AudioSplitter to the Splitter property to transcribe ` +
+                        `audio this size.`);
+                }
+                const pieces = await this.Splitter.Split(audio, SPLIT_TARGET_BYTES);
+                if (pieces.length === 0) {
+                    throw new Error('The configured AudioSplitter returned no pieces');
+                }
+                const transcripts = [];
+                for (const piece of pieces) {
+                    // A piece the splitter left oversized would fail at the API with a size
+                    // error naming neither the splitter nor which piece; say so here instead.
+                    if (piece.byteLength > MAX_UPLOAD_BYTES) {
+                        throw new Error(`The configured AudioSplitter produced a ${(piece.byteLength / (1024 * 1024)).toFixed(1)}MB ` +
+                            `piece, above Groq's 25MB limit`);
+                    }
+                    transcripts.push(await this.transcribeOne(piece, model, params));
+                }
+                text = transcripts.filter((t) => t.length > 0).join(' ');
+            }
+            if (text.trim().length === 0) {
+                // Whisper returns an empty string for silence, and for audio it could not
+                // decode. Both are failures from the caller's point of view — an empty
+                // transcript reported as success gets stored as if it were the content.
+                throw new Error('Transcription returned no text; the audio may be silent or in an unsupported format');
+            }
+            result.success = true;
+            result.content = text;
+        }
+        catch (error) {
+            const errorInfo = ErrorAnalyzer.analyzeError(error, 'Groq Whisper');
+            result.success = false;
+            result.errorMessage = error?.message || 'Unknown error occurred';
+            console.error('Groq Whisper error:', error, errorInfo);
+        }
+        return result;
+    }
+    /**
+     * Resolve the audio bytes from whichever of the two input forms the caller used.
+     */
+    resolveAudio(params) {
+        if (params.audioData && params.audioData.byteLength > 0) {
+            return params.audioData;
+        }
+        if (params.audioFile && params.audioFile.length > 0) {
+            return Buffer.from(params.audioFile, 'base64');
+        }
+        throw new Error('SpeechToText requires either audioData (Buffer) or audioFile (base 64 string)');
+    }
+    async transcribeOne(audio, model, params) {
+        const response = await this._client.audio.transcriptions.create({
+            file: await toFile(audio, params.fileName || 'audio.mp3'),
+            model,
+            response_format: 'json',
+            language: params.language,
+            prompt: params.prompt,
+            temperature: params.temperature,
+        });
+        return response.text ?? '';
+    }
+    async CreateSpeech(_params) {
+        throw new Error('Groq does not offer a text-to-speech API');
+    }
+    async GetVoices() {
+        // Transcription only — there are no voices to enumerate.
+        return [];
+    }
+    async GetModels() {
+        // Hardcoded rather than listed from the API: Groq's /models endpoint returns every
+        // model on the platform, LLMs included, with nothing on the row to identify which
+        // ones the transcription endpoint accepts.
+        return [
+            {
+                id: 'whisper-large-v3',
+                name: 'Whisper Large v3',
+                supportsTextToSpeech: false,
+                supportsVoiceConversion: false,
+                supportsStyle: false,
+                supportsSpeakerBoost: false,
+                supportsFineTuning: false,
+            },
+            {
+                id: 'whisper-large-v3-turbo',
+                name: 'Whisper Large v3 Turbo',
+                supportsTextToSpeech: false,
+                supportsVoiceConversion: false,
+                supportsStyle: false,
+                supportsSpeakerBoost: false,
+                supportsFineTuning: false,
+            },
+        ];
+    }
+    async GetPronounciationDictionaries() {
+        return [];
+    }
+    async GetSupportedMethods() {
+        return ['SpeechToText', 'GetModels'];
+    }
+};
+GroqAudioGenerator = __decorate([
+    RegisterClass(BaseAudioGenerator, 'GroqAudioGenerator'),
+    __metadata("design:paramtypes", [String])
+], GroqAudioGenerator);
+export { GroqAudioGenerator };
+//# sourceMappingURL=groqAudio.js.map

package/dist/models/groqAudio.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"groqAudio.js","sourceRoot":"","sources":["../../src/models/groqAudio.ts"],"names":[],"mappings":";;;;;;;;;AAAA,OAAO,EAGH,kBAAkB,EAClB,aAAa,EAEb,YAAY,GAIf,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EAAE,aAAa,EAAE,MAAM,wBAAwB,CAAC;AACvD,OAAO,IAAI,EAAE,EAAE,MAAM,EAAE,MAAM,UAAU,CAAC;AAExC;;;GAGG;AACH,MAAM,gBAAgB,GAAG,EAAE,GAAG,IAAI,GAAG,IAAI,CAAC;AAE1C;;;GAGG;AACH,MAAM,kBAAkB,GAAG,EAAE,GAAG,IAAI,GAAG,IAAI,CAAC;AAE5C,MAAM,aAAa,GAAG,kBAAkB,CAAC;AAEzC;;;;;;;;;;;;;GAaG;AAEI,IAAM,kBAAkB,GAAxB,MAAM,kBAAmB,SAAQ,kBAAkB;IAGtD,YAAY,MAAc;QACtB,KAAK,CAAC,MAAM,CAAC,CAAC;QAWlB;;;WAGG;QACI,aAAQ,GAAyB,IAAI,CAAC;QAdzC,IAAI,CAAC,OAAO,GAAG,IAAI,IAAI,CAAC,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;IAChD,CAAC;IAED;;OAEG;IACH,IAAW,UAAU;QACjB,OAAO,IAAI,CAAC,OAAO,CAAC;IACxB,CAAC;IAQD;;;;;;;OAOG;IACI,KAAK,CAAC,YAAY,CAAC,MAA0B;QAChD,MAAM,MAAM,GAAG,IAAI,YAAY,EAAE,CAAC;QAClC,IAAI,CAAC;YACD,MAAM,KAAK,GAAG,IAAI,CAAC,YAAY,CAAC,MAAM,CAAC,CAAC;YACxC,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,IAAI,aAAa,CAAC;YAE5C,IAAI,IAAY,CAAC;YACjB,IAAI,KAAK,CAAC,UAAU,IAAI,gBAAgB,EAAE,CAAC;gBACvC,IAAI,GAAG,MAAM,IAAI,CAAC,aAAa,CAAC,KAAK,EAAE,KAAK,EAAE,MAAM,CAAC,CAAC;YAC1D,CAAC;iBAAM,CAAC;gBACJ,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,CAAC;oBACjB,MAAM,IAAI,KAAK,CACX,YAAY,CAAC,KAAK,CAAC,UAAU,GAAG,CAAC,IAAI,GAAG,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,wBAAwB;wBAC7E,sFAAsF;wBACtF,kBAAkB,CACzB,CAAC;gBACN,CAAC;gBACD,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,KAAK,EAAE,kBAAkB,CAAC,CAAC;gBACpE,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;oBACtB,MAAM,IAAI,KAAK,CAAC,iDAAiD,CAAC,CAAC;gBACvE,CAAC;gBAED,MAAM,WAAW,GAAa,EAAE,CAAC;gBACjC,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;oBACzB,wEAAwE;oBACxE,0EAA0E;oBAC1E,IAAI,KAAK,CAAC,UAAU,GAAG,gBAAgB,EAAE,CAAC;wBACtC,MAAM,IAAI,KAAK,CACX,2CAA2C,CAAC,KAAK,CAAC,UAAU,GAAG,CAAC,IAAI,GAAG,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK;4BACzF,gCAAgC,CACvC,CAAC;oBACN,CAAC;oBACD,WAAW,CAAC,IAAI,CAAC,MAAM,IAAI,CAAC,aAAa,CAAC,KAAK,EAAE,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC;gBACrE,CAAC;gBACD,IAAI,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YAC7D,CAAC;YAED,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBAC3B,0EAA0E;gBAC1E,uEAAuE;gBACvE,wEAAwE;gBACxE,MAAM,IAAI,KAAK,CAAC,qFAAqF,CAAC,CAAC;YAC3G,CAAC;YAED,MAAM,CAAC,OAAO,GAAG,IAAI,CAAC;YACtB,MAAM,CAAC,OAAO,GAAG,IAAI,CAAC;QAC1B,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACb,MAAM,SAAS,GAAG,aAAa,CAAC,YAAY,CAAC,KAAK,EAAE,cAAc,CAAC,CAAC;YACpE,MAAM,CAAC,OAAO,GAAG,KAAK,CAAC;YACvB,MAAM,CAAC,YAAY,GAAG,KAAK,EAAE,OAAO,IAAI,wBAAwB,CAAC;YACjE,OAAO,CAAC,KAAK,CAAC,qBAAqB,EAAE,KAAK,EAAE,SAAS,CAAC,CAAC;QAC3D,CAAC;QACD,OAAO,MAAM,CAAC;IAClB,CAAC;IAED;;OAEG;IACK,YAAY,CAAC,MAA0B;QAC3C,IAAI,MAAM,CAAC,SAAS,IAAI,MAAM,CAAC,SAAS,CAAC,UAAU,GAAG,CAAC,EAAE,CAAC;YACtD,OAAO,MAAM,CAAC,SAAS,CAAC;QAC5B,CAAC;QACD,IAAI,MAAM,CAAC,SAAS,IAAI,MAAM,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAClD,OAAO,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAC;QACnD,CAAC;QACD,MAAM,IAAI,KAAK,CAAC,+EAA+E,CAAC,CAAC;IACrG,CAAC;IAEO,KAAK,CAAC,aAAa,CAAC,KAAa,EAAE,KAAa,EAAE,MAA0B;QAChF,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,cAAc,CAAC,MAAM,CAAC;YAC5D,IAAI,EAAE,MAAM,MAAM,CAAC,KAAK,EAAE,MAAM,CAAC,QAAQ,IAAI,WAAW,CAAC;YACzD,KAAK;YACL,eAAe,EAAE,MAAM;YACvB,QAAQ,EAAE,MAAM,CAAC,QAAQ;YACzB,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,WAAW,EAAE,MAAM,CAAC,WAAW;SAClC,CAAC,CAAC;QACH,OAAO,QAAQ,CAAC,IAAI,IAAI,EAAE,CAAC;IAC/B,CAAC;IAEM,KAAK,CAAC,YAAY,CAAC,OAA2B;QACjD,MAAM,IAAI,KAAK,CAAC,0CAA0C,CAAC,CAAC;IAChE,CAAC;IAEM,KAAK,CAAC,SAAS;QAClB,yDAAyD;QACzD,OAAO,EAAE,CAAC;IACd,CAAC;IAEM,KAAK,CAAC,SAAS;QAClB,mFAAmF;QACnF,kFAAkF;QAClF,2CAA2C;QAC3C,OAAO;YACH;gBACI,EAAE,EAAE,kBAAkB;gBACtB,IAAI,EAAE,kBAAkB;gBACxB,oBAAoB,EAAE,KAAK;gBAC3B,uBAAuB,EAAE,KAAK;gBAC9B,aAAa,EAAE,KAAK;gBACpB,oBAAoB,EAAE,KAAK;gBAC3B,kBAAkB,EAAE,KAAK;aAC5B;YACD;gBACI,EAAE,EAAE,wBAAwB;gBAC5B,IAAI,EAAE,wBAAwB;gBAC9B,oBAAoB,EAAE,KAAK;gBAC3B,uBAAuB,EAAE,KAAK;gBAC9B,aAAa,EAAE,KAAK;gBACpB,oBAAoB,EAAE,KAAK;gBAC3B,kBAAkB,EAAE,KAAK;aAC5B;SACJ,CAAC;IACN,CAAC;IAEM,KAAK,CAAC,6BAA6B;QACtC,OAAO,EAAE,CAAC;IACd,CAAC;IAEM,KAAK,CAAC,mBAAmB;QAC5B,OAAO,CAAC,cAAc,EAAE,WAAW,CAAC,CAAC;IACzC,CAAC;CACJ,CAAA;AAvJY,kBAAkB;IAD9B,aAAa,CAAC,kBAAkB,EAAE,oBAAoB,CAAC;;GAC3C,kBAAkB,CAuJ9B"}

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@memberjunction/ai-groq",
   "type": "module",
-  "version": "6.1.0-edge.0",
+  "version": "6.1.0-edge.2",
   "description": "MemberJunction Wrapper for Groq AI LPU inference engine",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
@@ -15,8 +15,8 @@
     "typescript": "^5.9.3"
   },
   "dependencies": {
-    "@memberjunction/ai": "6.1.0-edge.0",
-    "@memberjunction/global": "6.1.0-edge.0",
+    "@memberjunction/ai": "6.1.0-edge.2",
+    "@memberjunction/global": "6.1.0-edge.2",
     "groq-sdk": "^0.37.0"
   },
   "repository": {