speechflow 0.9.8 → 0.9.9

package/dst/speechflow-node-t2a-kokoro.js

@@ -0,0 +1,147 @@
+"use strict";
+/*
+**  SpeechFlow - Speech Processing Flow Graph
+**  Copyright (c) 2024-2025 Dr. Ralf S. Engelschall <rse@engelschall.com>
+**  Licensed under GPL 3.0 <https://spdx.org/licenses/GPL-3.0-only>
+*/
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+/*  standard dependencies  */
+const node_stream_1 = __importDefault(require("node:stream"));
+/*  external dependencies  */
+const kokoro_js_1 = require("kokoro-js");
+const speex_resampler_1 = __importDefault(require("speex-resampler"));
+/*  internal dependencies  */
+const speechflow_node_1 = __importDefault(require("./speechflow-node"));
+/*  SpeechFlow node for Kokoro text-to-speech conversion  */
+class SpeechFlowNodeKokoro extends speechflow_node_1.default {
+    /*  declare official node name  */
+    static name = "kokoro";
+    /*  internal state  */
+    kokoro = null;
+    static speexInitialized = false;
+    /*  construct node  */
+    constructor(id, cfg, opts, args) {
+        super(id, cfg, opts, args);
+        /*  declare node configuration parameters  */
+        this.configure({
+            voice: { type: "string", val: "Aoede", pos: 0, match: /^(?:Aoede|Heart|Puck|Fenrir)$/ },
+            language: { type: "string", val: "en", pos: 1, match: /^(?:en)$/ },
+            speed: { type: "number", val: 1.25, pos: 2, match: (n) => n >= 1.0 && n <= 1.30 },
+        });
+        /*  declare node input/output format  */
+        this.input = "text";
+        this.output = "audio";
+    }
+    /*  open node  */
+    async open() {
+        /*  establish Kokoro  */
+        const model = "onnx-community/Kokoro-82M-v1.0-ONNX";
+        const progressState = new Map();
+        const progressCallback = (progress) => {
+            let artifact = model;
+            if (typeof progress.file === "string")
+                artifact += `:${progress.file}`;
+            let percent = 0;
+            if (typeof progress.loaded === "number" && typeof progress.total === "number")
+                percent = progress.loaded / progress.total * 100;
+            else if (typeof progress.progress === "number")
+                percent = progress.progress;
+            if (percent > 0)
+                progressState.set(artifact, percent);
+        };
+        const interval = setInterval(() => {
+            for (const [artifact, percent] of progressState) {
+                this.log("info", `downloaded ${percent.toFixed(2)}% of artifact "${artifact}"`);
+                if (percent >= 1.0)
+                    progressState.delete(artifact);
+            }
+        }, 1000);
+        this.kokoro = await kokoro_js_1.KokoroTTS.from_pretrained(model, {
+            dtype: "q4f16",
+            progress_callback: progressCallback
+        });
+        clearInterval(interval);
+        if (this.kokoro === null)
+            throw new Error("failed to instantiate Kokoro");
+        /*  establish resampler from Kokoro's maximum 24Khz
+            output to our standard audio sample rate (48KHz)  */
+        if (!SpeechFlowNodeKokoro.speexInitialized) {
+            /*  at least once initialize resampler  */
+            await speex_resampler_1.default.initPromise;
+            SpeechFlowNodeKokoro.speexInitialized = true;
+        }
+        const resampler = new speex_resampler_1.default(1, 24000, this.config.audioSampleRate, 7);
+        /*  determine voice for text-to-speech operation  */
+        const voices = {
+            "Aoede": "af_aoede",
+            "Heart": "af_heart",
+            "Puck": "am_puck",
+            "Fenrir": "am_fenrir"
+        };
+        const voice = (voices[this.params.voice]);
+        if (voice === undefined)
+            throw new Error(`invalid Kokoro voice "${this.params.voice}"`);
+        /*  perform text-to-speech operation with Elevenlabs API  */
+        const text2speech = async (text) => {
+            this.log("info", `Kokoro: input: "${text}"`);
+            const audio = await this.kokoro.generate(text, {
+                speed: this.params.speed,
+                voice: voice
+            });
+            if (audio.sampling_rate !== 24000)
+                throw new Error("expected 24KHz sampling rate in Kokoro output");
+            /*  convert audio samples from PCM/F32/24Khz to PCM/I16/24KHz  */
+            const samples = audio.audio;
+            const buffer1 = Buffer.alloc(samples.length * 2);
+            for (let i = 0; i < samples.length; i++) {
+                const sample = Math.max(-1, Math.min(1, samples[i]));
+                buffer1.writeInt16LE(sample * 0x7FFF, i * 2);
+            }
+            /*  resample audio samples from PCM/I16/24Khz to PCM/I16/48KHz  */
+            const buffer2 = resampler.processChunk(buffer1);
+            return buffer2;
+        };
+        /*  create transform stream and connect it to the Kokoro API  */
+        const log = (level, msg) => { this.log(level, msg); };
+        this.stream = new node_stream_1.default.Transform({
+            writableObjectMode: true,
+            readableObjectMode: true,
+            decodeStrings: false,
+            transform(chunk, encoding, callback) {
+                if (Buffer.isBuffer(chunk.payload))
+                    callback(new Error("invalid chunk payload type"));
+                else {
+                    text2speech(chunk.payload).then((buffer) => {
+                        log("info", `Kokoro: received audio (buffer length: ${buffer.byteLength})`);
+                        chunk = chunk.clone();
+                        chunk.type = "audio";
+                        chunk.payload = buffer;
+                        this.push(chunk);
+                        callback();
+                    }).catch((err) => {
+                        callback(err);
+                    });
+                }
+            },
+            final(callback) {
+                this.push(null);
+                callback();
+            }
+        });
+    }
+    /*  close node  */
+    async close() {
+        /*  destroy stream  */
+        if (this.stream !== null) {
+            this.stream.destroy();
+            this.stream = null;
+        }
+        /*  destroy Kokoro API  */
+        if (this.kokoro !== null)
+            this.kokoro = null;
+    }
+}
+exports.default = SpeechFlowNodeKokoro;

package/dst/speechflow-node-t2t-gemma.js

@@ -93,7 +93,8 @@ class SpeechFlowNodeGemma extends speechflow_node_1.default {
                 "Do not show any prolog.\n" +
                 "Do not show any epilog.\n" +
                 "Get to the point.\n" +
-                "Directly translate text from Enlish (EN) to German (DE) language.\n",
+                "Preserve the original meaning, tone, and nuance.\n" +
+                "Directly translate text from English (EN) to fluent and natural German (DE) language.\n",
             chat: [
                 { role: "user", content: "I love my wife." },
                 { role: "system", content: "Ich liebe meine Frau." },
@@ -115,7 +116,8 @@ class SpeechFlowNodeGemma extends speechflow_node_1.default {
                 "Do not show any prolog. \n" +
                 "Do not show any epilog. \n" +
                 "Get to the point.\n" +
-                "Directly translate text from German (DE) to English (EN) language.\n",
+                "Preserve the original meaning, tone, and nuance.\n" +
+                "Directly translate text from German (DE) to fluent and natural English (EN) language.\n",
             chat: [
                 { role: "user", content: "Ich liebe meine Frau." },
                 { role: "system", content: "I love my wife." },
@@ -132,9 +134,16 @@ class SpeechFlowNodeGemma extends speechflow_node_1.default {
         /*  declare node configuration parameters  */
         this.configure({
             api: { type: "string", val: "http://127.0.0.1:11434", match: /^https?:\/\/.+?:\d+$/ },
+            model: { type: "string", val: "gemma3:4b-it-q4_K_M", match: /^.+$/ },
             src: { type: "string", pos: 0, val: "de", match: /^(?:de|en)$/ },
             dst: { type: "string", pos: 1, val: "en", match: /^(?:de|en)$/ }
         });
+        /*  tell effective mode  */
+        if (this.params.src === this.params.dst)
+            this.log("info", `Gemma/Ollama: operation mode: spellchecking for language "${this.params.src}"`);
+        else
+            this.log("info", `Gemma/Ollama: operation mode: translation from language "${this.params.src}"` +
+                ` to language "${this.params.dst}"`);
         /*  declare node input/output format  */
         this.input = "text";
         this.output = "text";
@@ -143,12 +152,23 @@ class SpeechFlowNodeGemma extends speechflow_node_1.default {
     async open() {
         /*  instantiate Ollama API  */
         this.ollama = new ollama_1.Ollama({ host: this.params.api });
+        /*  ensure the model is available  */
+        const model = this.params.model;
+        const models = await this.ollama.list();
+        const exists = models.models.some((m) => m.name === model);
+        if (!exists) {
+            this.log("info", `Gemma/Ollama: model "${model}" still not present in Ollama -- ` +
+                "automatically downloading model");
+            await this.ollama.pull({ model });
+        }
+        else
+            this.log("info", `Gemma/Ollama: model "${model}" already present in Ollama`);
         /*  provide text-to-text translation  */
         const translate = async (text) => {
             const key = `${this.params.src}-${this.params.dst}`;
             const cfg = this.setup[key];
             const response = await this.ollama.chat({
-                model: "gemma3:4b-it-q4_K_M",
+                model,
                 messages: [
                     { role: "system", content: cfg.systemPrompt },
                     ...cfg.chat,

package/dst/speechflow-node-t2t-ollama.d.ts

@@ -0,0 +1,13 @@
+import SpeechFlowNode from "./speechflow-node";
+export default class SpeechFlowNodeOllama extends SpeechFlowNode {
+    static name: string;
+    private ollama;
+    private setup;
+    constructor(id: string, cfg: {
+        [id: string]: any;
+    }, opts: {
+        [id: string]: any;
+    }, args: any[]);
+    open(): Promise<void>;
+    close(): Promise<void>;
+}

package/dst/speechflow-node-t2t-ollama.js

@@ -0,0 +1,245 @@
+"use strict";
+/*
+**  SpeechFlow - Speech Processing Flow Graph
+**  Copyright (c) 2024-2025 Dr. Ralf S. Engelschall <rse@engelschall.com>
+**  Licensed under GPL 3.0 <https://spdx.org/licenses/GPL-3.0-only>
+*/
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
+Object.defineProperty(exports, "__esModule", { value: true });
+/*  standard dependencies  */
+const node_stream_1 = __importDefault(require("node:stream"));
+/*  external dependencies  */
+const ollama_1 = require("ollama");
+/*  internal dependencies  */
+const speechflow_node_1 = __importDefault(require("./speechflow-node"));
+/*  SpeechFlow node for Ollama text-to-text translation  */
+class SpeechFlowNodeOllama extends speechflow_node_1.default {
+    /*  declare official node name  */
+    static name = "ollama";
+    /*  internal state  */
+    ollama = null;
+    /*  internal LLM setup  */
+    setup = {
+        /*  English (EN) spellchecking only  */
+        "en-en": {
+            systemPrompt: "You are a proofreader and spellchecker for English.\n" +
+                "Output only the corrected text.\n" +
+                "Do NOT use markdown.\n" +
+                "Do NOT give any explanations.\n" +
+                "Do NOT give any introduction.\n" +
+                "Do NOT give any comments.\n" +
+                "Do NOT give any preamble.\n" +
+                "Do NOT give any prolog.\n" +
+                "Do NOT give any epilog.\n" +
+                "Do NOT change the gammar.\n" +
+                "Do NOT use synonyms for words.\n" +
+                "Keep all words.\n" +
+                "Fill in missing commas.\n" +
+                "Fill in missing points.\n" +
+                "Fill in missing question marks.\n" +
+                "Fill in missing hyphens.\n" +
+                "Focus ONLY on the word spelling.\n" +
+                "The text you have to correct is:\n",
+            chat: [
+                { role: "user", content: "I luve my wyfe" },
+                { role: "system", content: "I love my wife." },
+                { role: "user", content: "The weether is wunderfull!" },
+                { role: "system", content: "The weather is wonderful!" },
+                { role: "user", content: "The live awesome but I'm hungry." },
+                { role: "system", content: "The live is awesome, but I'm hungry." }
+            ]
+        },
+        /*  German (DE) spellchecking only  */
+        "de-de": {
+            systemPrompt: "Du bist ein Korrekturleser und Rechtschreibprüfer für Deutsch.\n" +
+                "Gib nur den korrigierten Text aus.\n" +
+                "Benutze KEIN Markdown.\n" +
+                "Gib KEINE Erklärungen.\n" +
+                "Gib KEINE Einleitung.\n" +
+                "Gib KEINE Kommentare.\n" +
+                "Gib KEINE Preamble.\n" +
+                "Gib KEINEN Prolog.\n" +
+                "Gib KEINEN Epilog.\n" +
+                "Ändere NICHT die Grammatik.\n" +
+                "Verwende KEINE Synonyme für Wörter.\n" +
+                "Behalte alle Wörter bei.\n" +
+                "Füge fehlende Kommas ein.\n" +
+                "Füge fehlende Punkte ein.\n" +
+                "Füge fehlende Fragezeichen ein.\n" +
+                "Füge fehlende Bindestriche ein.\n" +
+                "Füge fehlende Gedankenstriche ein.\n" +
+                "Fokussiere dich NUR auf die Rechtschreibung der Wörter.\n" +
+                "Der von dir zu korrigierende Text ist:\n",
+            chat: [
+                { role: "user", content: "Ich ljebe meine Frao" },
+                { role: "system", content: "Ich liebe meine Frau." },
+                { role: "user", content: "Die Wedter ist wunderschoen." },
+                { role: "system", content: "Das Wetter ist wunderschön." },
+                { role: "user", content: "Das Leben einfach großartig aber ich bin hungrig." },
+                { role: "system", content: "Das Leben ist einfach großartig, aber ich bin hungrig." }
+            ]
+        },
+        /*  English (EN) to German (DE) translation  */
+        "en-de": {
+            systemPrompt: "You are a translator.\n" +
+                "Output only the requested text.\n" +
+                "Do not use markdown.\n" +
+                "Do not chat.\n" +
+                "Do not show any explanations.\n" +
+                "Do not show any introduction.\n" +
+                "Do not show any preamble.\n" +
+                "Do not show any prolog.\n" +
+                "Do not show any epilog.\n" +
+                "Get to the point.\n" +
+                "Preserve the original meaning, tone, and nuance.\n" +
+                "Directly translate text from English (EN) to fluent and natural German (DE) language.\n",
+            chat: [
+                { role: "user", content: "I love my wife." },
+                { role: "system", content: "Ich liebe meine Frau." },
+                { role: "user", content: "The weather is wonderful." },
+                { role: "system", content: "Das Wetter ist wunderschön." },
+                { role: "user", content: "The live is awesome." },
+                { role: "system", content: "Das Leben ist einfach großartig." }
+            ]
+        },
+        /*  German (DE) to English (EN) translation  */
+        "de-en": {
+            systemPrompt: "You are a translator.\n" +
+                "Output only the requested text.\n" +
+                "Do not use markdown.\n" +
+                "Do not chat.\n" +
+                "Do not show any explanations.\n" +
+                "Do not show any introduction.\n" +
+                "Do not show any preamble. \n" +
+                "Do not show any prolog. \n" +
+                "Do not show any epilog. \n" +
+                "Get to the point.\n" +
+                "Preserve the original meaning, tone, and nuance.\n" +
+                "Directly translate text from German (DE) to fluent and natural English (EN) language.\n",
+            chat: [
+                { role: "user", content: "Ich liebe meine Frau." },
+                { role: "system", content: "I love my wife." },
+                { role: "user", content: "Das Wetter ist wunderschön." },
+                { role: "system", content: "The weather is wonderful." },
+                { role: "user", content: "Das Leben ist einfach großartig." },
+                { role: "system", content: "The live is awesome." }
+            ]
+        }
+    };
+    /*  construct node  */
+    constructor(id, cfg, opts, args) {
+        super(id, cfg, opts, args);
+        /*  declare node configuration parameters  */
+        this.configure({
+            api: { type: "string", val: "http://127.0.0.1:11434", match: /^https?:\/\/.+?:\d+$/ },
+            model: { type: "string", val: "gemma3:4b-it-q4_K_M", match: /^.+$/ },
+            src: { type: "string", pos: 0, val: "de", match: /^(?:de|en)$/ },
+            dst: { type: "string", pos: 1, val: "en", match: /^(?:de|en)$/ }
+        });
+        /*  tell effective mode  */
+        if (this.params.src === this.params.dst)
+            this.log("info", `Ollama: operation mode: spellchecking for language "${this.params.src}"`);
+        else
+            this.log("info", `Ollama: operation mode: translation from language "${this.params.src}"` +
+                ` to language "${this.params.dst}"`);
+        /*  declare node input/output format  */
+        this.input = "text";
+        this.output = "text";
+    }
+    /*  open node  */
+    async open() {
+        /*  instantiate Ollama API  */
+        this.ollama = new ollama_1.Ollama({ host: this.params.api });
+        /*  ensure the model is available  */
+        const model = this.params.model;
+        const models = await this.ollama.list();
+        const exists = models.models.some((m) => m.name === model);
+        if (!exists) {
+            this.log("info", `Ollama: model "${model}" still not present in Ollama -- ` +
+                "automatically downloading model");
+            let artifact = "";
+            let percent = 0;
+            const interval = setInterval(() => {
+                this.log("info", `downloaded ${percent.toFixed(2)}% of artifact "${artifact}"`);
+            }, 1000);
+            const progress = await this.ollama.pull({ model, stream: true });
+            for await (const event of progress) {
+                if (event.digest)
+                    artifact = event.digest;
+                if (event.completed && event.total)
+                    percent = (event.completed / event.total) * 100;
+            }
+            clearInterval(interval);
+        }
+        else
+            this.log("info", `Ollama: model "${model}" already present in Ollama`);
+        /*  provide text-to-text translation  */
+        const translate = async (text) => {
+            const key = `${this.params.src}-${this.params.dst}`;
+            const cfg = this.setup[key];
+            const response = await this.ollama.chat({
+                model,
+                messages: [
+                    { role: "system", content: cfg.systemPrompt },
+                    ...cfg.chat,
+                    { role: "user", content: text }
+                ],
+                keep_alive: "10m",
+                options: {
+                    repeat_penalty: 1.1,
+                    temperature: 0.7,
+                    seed: 1,
+                    top_k: 10,
+                    top_p: 0.5
+                }
+            });
+            return response.message.content;
+        };
+        /*  establish a duplex stream and connect it to Ollama  */
+        this.stream = new node_stream_1.default.Transform({
+            readableObjectMode: true,
+            writableObjectMode: true,
+            decodeStrings: false,
+            transform(chunk, encoding, callback) {
+                if (Buffer.isBuffer(chunk.payload))
+                    callback(new Error("invalid chunk payload type"));
+                else {
+                    if (chunk.payload === "") {
+                        this.push(chunk);
+                        callback();
+                    }
+                    else {
+                        translate(chunk.payload).then((payload) => {
+                            const chunkNew = chunk.clone();
+                            chunkNew.payload = payload;
+                            this.push(chunkNew);
+                            callback();
+                        }).catch((err) => {
+                            callback(err);
+                        });
+                    }
+                }
+            },
+            final(callback) {
+                this.push(null);
+                callback();
+            }
+        });
+    }
+    /*  close node  */
+    async close() {
+        /*  close stream  */
+        if (this.stream !== null) {
+            this.stream.destroy();
+            this.stream = null;
+        }
+        /*  shutdown Ollama  */
+        if (this.ollama !== null) {
+            this.ollama.abort();
+            this.ollama = null;
+        }
+    }
+}
+exports.default = SpeechFlowNodeOllama;

package/dst/speechflow-node-t2t-openai.d.ts

@@ -0,0 +1,13 @@
+import SpeechFlowNode from "./speechflow-node";
+export default class SpeechFlowNodeOpenAI extends SpeechFlowNode {
+    static name: string;
+    private openai;
+    private setup;
+    constructor(id: string, cfg: {
+        [id: string]: any;
+    }, opts: {
+        [id: string]: any;
+    }, args: any[]);
+    open(): Promise<void>;
+    close(): Promise<void>;
+}