speechflow 1.7.0 → 2.0.0

package/speechflow-cli/src/speechflow-node-t2t-subtitle.ts

@@ -32,7 +32,7 @@ type TextChunk = {
     text:  string
 }
 
-/*  SpeechFlow node for subtitle (text-to-text) "translations"  */
+/*  SpeechFlow node for subtitle (text-to-text) conversions  */
 export default class SpeechFlowNodeT2TSubtitle extends SpeechFlowNode {
     /*  declare official node name  */
     public static name = "t2t-subtitle"
@@ -311,7 +311,7 @@ export default class SpeechFlowNodeT2TSubtitle extends SpeechFlowNode {
                 final (callback) {
                     /*  process any remaining buffer content  */
                     if (buffer.trim() !== "") {
-                        try {
+                        util.shield(() => {
                             /*  parse entries  */
                             const entries = self.params.format === "srt" ? parseSRT(buffer) : parseVTT(buffer)
 
@@ -320,10 +320,7 @@ export default class SpeechFlowNodeT2TSubtitle extends SpeechFlowNode {
                                 const chunkNew = new SpeechFlowChunk(entry.start, entry.end, "final", "text", entry.text)
                                 this.push(chunkNew)
                             }
-                        }
-                        catch (_error: unknown) {
-                            /*  ignore parse errors on final flush  */
-                        }
+                        })
                     }
                     callback()
                 }

package/speechflow-cli/src/speechflow-node-t2t-summary.ts

@@ -0,0 +1,229 @@
+/*
+**  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>
+*/
+
+/*  standard dependencies  */
+import Stream                              from "node:stream"
+
+/*  internal dependencies  */
+import SpeechFlowNode, { SpeechFlowChunk } from "./speechflow-node"
+import * as util                           from "./speechflow-util"
+import { LLM, type LLMCompleteMessage }    from "./speechflow-util-llm"
+
+/*  internal utility types  */
+type ConfigEntry = { systemPrompt: string, chat: LLMCompleteMessage[] }
+type Config      = { [ key: string ]: ConfigEntry }
+
+/*  SpeechFlow node for text-to-text summarization  */
+export default class SpeechFlowNodeT2TSummary extends SpeechFlowNode {
+    /*  declare official node name  */
+    public static name = "t2t-summary"
+
+    /*  internal state  */
+    private llm: LLM | null = null
+    private accumulatedText = ""
+    private sentencesSinceLastSummary = 0
+
+    /*  internal LLM setup  */
+    private setup: Config = {
+        /*  English (EN) summarization  */
+        "en": {
+            systemPrompt:
+                "You are a text summarizer.\n" +
+                "Output only the summary.\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" +
+                "Get to the point.\n" +
+                "Summarize the following text into %N% sentences.\n" +
+                "The text is:\n",
+            chat: [
+                { role: "user",      content: "The weather today is sunny and warm. Birds are singing in the trees. People are enjoying the outdoors." },
+                { role: "assistant", content: "The weather is pleasant with sunshine, birdsong, and people outdoors." },
+                { role: "user",      content: "John went to the store to buy groceries. He needed milk, bread, and eggs. The store was crowded but he found everything he needed." },
+                { role: "assistant", content: "John successfully bought milk, bread, and eggs from a crowded store." }
+            ]
+        },
+
+        /*  German (DE) summarization  */
+        "de": {
+            systemPrompt:
+                "Du bist ein Textzusammenfasser.\n" +
+                "Gib nur die Zusammenfassung aus.\n" +
+                "Benutze KEIN Markdown.\n" +
+                "Gib KEINE Erklärungen.\n" +
+                "Gib KEINE Einleitung.\n" +
+                "Gib KEINE Kommentare.\n" +
+                "Gib KEINE Prämbel.\n" +
+                "Gib KEINEN Prolog.\n" +
+                "Gib KEINEN Epilog.\n" +
+                "Komme auf den Punkt.\n" +
+                "Fasse den folgenden Text in %N% Sätzen zusammen.\n" +
+                "Der Text ist:\n",
+            chat: [
+                { role: "user",      content: "Das Wetter heute ist sonnig und warm. Vögel singen in den Bäumen. Die Menschen genießen die Zeit im Freien." },
+                { role: "assistant", content: "Das Wetter ist angenehm mit Sonnenschein, Vogelgesang und Menschen im Freien." },
+                { role: "user",      content: "Hans ging in den Laden, um Lebensmittel zu kaufen. Er brauchte Milch, Brot und Eier. Der Laden war voll, aber er fand alles, was er brauchte." },
+                { role: "assistant", content: "Hans kaufte erfolgreich Milch, Brot und Eier in einem vollen Laden." }
+            ]
+        }
+    }
+
+    /*  construct node  */
+    constructor (id: string, cfg: { [ id: string ]: any }, opts: { [ id: string ]: any }, args: any[]) {
+        super(id, cfg, opts, args)
+
+        /*  declare node configuration parameters  */
+        this.configure({
+            provider: { type: "string",         val: "ollama",                 match: /^(?:openai|anthropic|google|ollama|transformers)$/ },
+            api:      { type: "string",         val: "http://127.0.0.1:11434", match: /^https?:\/\/.+?(:\d+)?$/ },
+            model:    { type: "string",         val: "gemma3:4b-it-q4_K_M",    match: /^.+$/ },
+            key:      { type: "string",         val: "",                       match: /^.*$/ },
+            lang:     { type: "string", pos: 0, val: "en",                     match: /^(?:en|de)$/ },
+            size:     { type: "number", pos: 1, val: 4,                        match: (n: number) => n >= 1 && n <= 20 },
+            trigger:  { type: "number", pos: 2, val: 8,                        match: (n: number) => n >= 1 && n <= 100 }
+        })
+
+        /*  tell effective mode  */
+        this.log("info", `summarizing language "${this.params.lang}" ` +
+            `via ${this.params.provider} LLM (model: ${this.params.model}), ` +
+            `triggering every new ${this.params.trigger} sentences, ` +
+            `summarizing into ${this.params.size} sentences`)
+
+        /*  declare node input/output format  */
+        this.input  = "text"
+        this.output = "text"
+    }
+
+    /*  count sentences in text  */
+    private countSentences (text: string): number {
+        const matches = text.match(/[.;?!]/g)
+        return matches ? matches.length : 0
+    }
+
+    /*  open node  */
+    async open () {
+        /*  reset internal state  */
+        this.accumulatedText = ""
+        this.sentencesSinceLastSummary = 0
+
+        /*  instantiate LLM  */
+        this.llm = new LLM({
+            provider:    this.params.provider,
+            api:         this.params.api,
+            model:       this.params.model,
+            key:         this.params.key,
+            temperature: 0.7,
+            topP:        0.5
+        })
+        this.llm.on("log", (level: string, message: string) => {
+            this.log(level as "info" | "warning" | "error", message)
+        })
+        await this.llm.open()
+
+        /*  provide text summarization  */
+        const llm = this.llm!
+        const summarize = async (text: string) => {
+            const cfg = this.setup[this.params.lang]
+            if (!cfg)
+                throw new Error(`unsupported language: ${this.params.lang}`)
+            return llm.complete({
+                system:   cfg.systemPrompt.replace(/%N%/, this.params.size),
+                messages: cfg.chat,
+                prompt:   text
+            })
+        }
+
+        /*  establish a transform stream for summarization  */
+        const self = this
+        this.stream = new Stream.Transform({
+            readableObjectMode: true,
+            writableObjectMode: true,
+            decodeStrings:      false,
+            highWaterMark:      1,
+            transform (chunk: SpeechFlowChunk, encoding, callback) {
+                if (Buffer.isBuffer(chunk.payload))
+                    callback(new Error("invalid chunk payload type"))
+                else if (chunk.payload === "") {
+                    this.push(chunk)
+                    callback()
+                }
+                else {
+                    /*  accumulate text  */
+                    if (self.accumulatedText.length > 0)
+                        self.accumulatedText += " "
+                    self.accumulatedText += chunk.payload
+
+                    /*  count new sentences  */
+                    const newSentences = self.countSentences(chunk.payload)
+                    self.sentencesSinceLastSummary += newSentences
+                    self.log("info", `accumulated ${self.sentencesSinceLastSummary} sentences ` +
+                        `(trigger: ${self.params.trigger})`)
+
+                    /*  check if we should generate a summary  */
+                    if (self.sentencesSinceLastSummary >= self.params.trigger) {
+                        self.sentencesSinceLastSummary = 0
+                        self.log("info", `generating summary of accumulated text`)
+                        const textToSummarize = self.accumulatedText
+                        self.accumulatedText = ""
+                        summarize(textToSummarize).then((summary) => {
+                            const chunkNew = chunk.clone()
+                            chunkNew.payload = summary
+                            this.push(chunkNew)
+                            callback()
+                        }).catch((error: unknown) => {
+                            callback(util.ensureError(error))
+                        })
+                    }
+                    else
+                        callback()
+                }
+            },
+            final (callback) {
+                /*  generate final summary if there is accumulated text  */
+                if (self.accumulatedText.length > 0 && self.sentencesSinceLastSummary > 0) {
+                    self.sentencesSinceLastSummary = 0
+                    self.log("info", `generating final summary of accumulated text`)
+                    const textToSummarize = self.accumulatedText
+                    self.accumulatedText = ""
+                    summarize(textToSummarize).then((summary) => {
+                        const chunkNew = new SpeechFlowChunk(
+                            self.timeZeroOffset, self.timeZeroOffset,
+                            "final", "text", summary)
+                        this.push(chunkNew)
+                        callback()
+                    }).catch((error: unknown) => {
+                        callback(util.ensureError(error))
+                    })
+                }
+                else
+                    callback()
+            }
+        })
+    }
+
+    /*  close node  */
+    async close () {
+        /*  reset internal state  */
+        this.accumulatedText = ""
+        this.sentencesSinceLastSummary = 0
+
+        /*  shutdown stream  */
+        if (this.stream !== null) {
+            await util.destroyStream(this.stream)
+            this.stream = null
+        }
+
+        /*  shutdown LLM  */
+        if (this.llm !== null) {
+            await this.llm.close()
+            this.llm = null
+        }
+    }
+}

package/speechflow-cli/src/speechflow-node-t2t-translate.ts

@@ -0,0 +1,181 @@
+/*
+**  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>
+*/
+
+/*  standard dependencies  */
+import Stream                              from "node:stream"
+
+/*  internal dependencies  */
+import SpeechFlowNode, { SpeechFlowChunk } from "./speechflow-node"
+import * as util                           from "./speechflow-util"
+import { LLM, type LLMCompleteMessage }    from "./speechflow-util-llm"
+
+/*  internal utility types  */
+type ConfigEntry = { systemPrompt: string, chat: LLMCompleteMessage[] }
+type Config      = { [ key: string ]: ConfigEntry }
+
+/*  SpeechFlow node for LLM-based text-to-text translation  */
+export default class SpeechFlowNodeT2TTranslate extends SpeechFlowNode {
+    /*  declare official node name  */
+    public static name = "t2t-translate"
+
+    /*  internal state  */
+    private llm: LLM | null = null
+
+    /*  internal LLM setup  */
+    private setup: Config = {
+        /*  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: "assistant", content: "Ich liebe meine Frau." },
+                { role: "user",      content: "The weather is wonderful." },
+                { role: "assistant", content: "Das Wetter ist wunderschön." },
+                { role: "user",      content: "The life is awesome." },
+                { role: "assistant", 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: "assistant", content: "I love my wife." },
+                { role: "user",      content: "Das Wetter ist wunderschön." },
+                { role: "assistant", content: "The weather is wonderful." },
+                { role: "user",      content: "Das Leben ist einfach großartig." },
+                { role: "assistant", content: "The life is awesome." }
+            ]
+        }
+    }
+
+    /*  construct node  */
+    constructor (id: string, cfg: { [ id: string ]: any }, opts: { [ id: string ]: any }, args: any[]) {
+        super(id, cfg, opts, args)
+
+        /*  declare node configuration parameters  */
+        this.configure({
+            src:      { type: "string", pos: 0, val: "de",                     match: /^(?:de|en)$/ },
+            dst:      { type: "string", pos: 1, val: "en",                     match: /^(?:de|en)$/ },
+            provider: { type: "string",         val: "ollama",                 match: /^(?:openai|anthropic|google|ollama|transformers)$/ },
+            api:      { type: "string",         val: "http://127.0.0.1:11434", match: /^https?:\/\/.+?(:\d+)?$/ },
+            model:    { type: "string",         val: "gemma3:4b-it-q4_K_M",    match: /^.+$/ },
+            key:      { type: "string",         val: "",                       match: /^.*$/ }
+        })
+
+        /*  validate translation direction  */
+        if (this.params.src === this.params.dst)
+            throw new Error("source and destination language must be different for translation")
+
+        /*  tell effective mode  */
+        this.log("info", `translating from language "${this.params.src}" to language "${this.params.dst}" ` +
+            `via ${this.params.provider} LLM (model: ${this.params.model})`)
+
+        /*  declare node input/output format  */
+        this.input  = "text"
+        this.output = "text"
+    }
+
+    /*  open node  */
+    async open () {
+        /*  instantiate LLM  */
+        this.llm = new LLM({
+            provider:    this.params.provider,
+            api:         this.params.api,
+            model:       this.params.model,
+            key:         this.params.key,
+            temperature: 0.7,
+            topP:        0.5
+        })
+        this.llm.on("log", (level: string, message: string) => {
+            this.log(level as "info" | "warning" | "error", message)
+        })
+        await this.llm.open()
+
+        /*  provide text-to-text translation  */
+        const llm = this.llm!
+        const translate = async (text: string) => {
+            const key = `${this.params.src}-${this.params.dst}`
+            const cfg = this.setup[key]
+            if (!cfg)
+                throw new Error(`unsupported language pair: ${key}`)
+            return llm.complete({
+                system:   cfg.systemPrompt,
+                messages: cfg.chat,
+                prompt:   text
+            })
+        }
+
+        /*  establish a transform stream and connect it to LLM  */
+        this.stream = new Stream.Transform({
+            readableObjectMode: true,
+            writableObjectMode: true,
+            decodeStrings:      false,
+            highWaterMark:      1,
+            transform (chunk: SpeechFlowChunk, 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((error: unknown) => {
+                        callback(util.ensureError(error))
+                    })
+                }
+            },
+            final (callback) {
+                callback()
+            }
+        })
+    }
+
+    /*  close node  */
+    async close () {
+        /*  shutdown stream  */
+        if (this.stream !== null) {
+            await util.destroyStream(this.stream)
+            this.stream = null
+        }
+
+        /*  shutdown LLM  */
+        if (this.llm !== null) {
+            await this.llm.close()
+            this.llm = null
+        }
+    }
+}

package/speechflow-cli/src/speechflow-node-x2x-filter.ts

@@ -19,6 +19,9 @@ export default class SpeechFlowNodeX2XFilter extends SpeechFlowNode {
     /*  cached regular expression instance  */
     private cachedRegExp = new util.CachedRegExp()
 
+    /*  internal state  */
+    private closing = false
+
     /*  construct node  */
     constructor (id: string, cfg: { [ id: string ]: any }, opts: { [ id: string ]: any }, args: any[]) {
         super(id, cfg, opts, args)
@@ -39,17 +42,20 @@ export default class SpeechFlowNodeX2XFilter extends SpeechFlowNode {
 
     /*  open node  */
     async open () {
+        /*  clear destruction flag  */
+        this.closing = false
+
         /*  helper function for comparing two values  */
         const comparison = (val1: any, op: string, val2: any) => {
             if (op === "==" || op === "!=") {
                 /*  equal comparison  */
-                const str1 = (typeof val1 === "string" ? val1 : val1.toString()) as string
-                const str2 = (typeof val2 === "string" ? val2 : val2.toString()) as string
+                const str1 = (typeof val1 === "string" ? val1 : val1.toString())
+                const str2 = (typeof val2 === "string" ? val2 : val2.toString())
                 return (op === "==" ? (str1 === str2) : (str1 !== str2))
             }
             else if (op === "~~" || op === "!~") {
                 /*  regular expression comparison  */
-                const str = (typeof val1 === "string" ? val1 : val1.toString()) as string
+                const str = (typeof val1 === "string" ? val1 : val1.toString())
                 const regexp = (
                     val2 instanceof RegExp ?
                         val2 :
@@ -93,6 +99,10 @@ export default class SpeechFlowNodeX2XFilter extends SpeechFlowNode {
             decodeStrings:      false,
             highWaterMark:      1,
             transform (chunk: SpeechFlowChunk, encoding, callback) {
+                if (self.closing) {
+                    callback(new Error("stream already destroyed"))
+                    return
+                }
                 let val1: any
                 const val2: any = self.params.val
                 const m = self.params.var.match(/^meta:(.+)$/)
@@ -124,6 +134,9 @@ export default class SpeechFlowNodeX2XFilter extends SpeechFlowNode {
 
     /*  close node  */
     async close () {
+        /*  indicate closing  */
+        this.closing = true
+
         /*  shutdown stream  */
         if (this.stream !== null) {
             await util.destroyStream(this.stream)

package/speechflow-cli/src/speechflow-node-x2x-trace.ts

@@ -28,10 +28,10 @@ export default class SpeechFlowNodeX2XTrace extends SpeechFlowNode {
 
         /*  declare node configuration parameters  */
         this.configure({
-            type:      { type: "string", pos: 0, val: "audio", match: /^(?:audio|text)$/ },
-            name:      { type: "string", pos: 1, val: "trace" },
+            type:      { type: "string", pos: 0, val: "audio",  match: /^(?:audio|text)$/  },
+            name:      { type: "string", pos: 1, val: "trace"                              },
             mode:      { type: "string", pos: 2, val: "filter", match: /^(?:filter|sink)$/ },
-            dashboard: { type: "string",         val: "" }
+            dashboard: { type: "string",         val: ""                                   }
         })
 
         /*  sanity check parameters  */

package/speechflow-cli/src/speechflow-node-xio-device.ts

@@ -20,10 +20,7 @@ export default class SpeechFlowNodeXIODevice extends SpeechFlowNode {
     public static name = "xio-device"
 
     /*  internal state  */
-    private io: PortAudio.IoStreamRead
-        | PortAudio.IoStreamWrite
-        | PortAudio.IoStreamDuplex
-        | null = null
+    private io: PortAudio.IoStreamRead | PortAudio.IoStreamWrite | PortAudio.IoStreamDuplex | null = null
 
     /*  construct node  */
     constructor (id: string, cfg: { [ id: string ]: any }, opts: { [ id: string ]: any }, args: any[]) {
@@ -87,7 +84,7 @@ export default class SpeechFlowNodeXIODevice extends SpeechFlowNode {
         return device
     }
 
-    /*  NOTICE: "naudion" actually implements Stream.{Readable,Writable,Duplex}, but
+    /*  NOTICE: "naudiodon" actually implements Stream.{Readable,Writable,Duplex}, but
         declares just its sub-interface NodeJS.{Readable,Writable,Duplex}Stream,
         so it is correct to cast it back to Stream.{Readable,Writable,Duplex}
         in the following device stream setup functions!  */
@@ -211,7 +208,7 @@ export default class SpeechFlowNodeXIODevice extends SpeechFlowNode {
                     throw error
             }
             await Promise.race([
-                util.timeoutPromise(2 * 1000, "PortAudio abort timeout"),
+                util.timeout(2 * 1000, "PortAudio abort timeout"),
                 new Promise<void>((resolve) => {
                     this.io!.abort(() => {
                         resolve()
@@ -219,7 +216,7 @@ export default class SpeechFlowNodeXIODevice extends SpeechFlowNode {
                 }).catch(catchHandler)
             ])
             await Promise.race([
-                util.timeoutPromise(2 * 1000, "PortAudio quit timeout"),
+                util.timeout(2 * 1000, "PortAudio quit timeout"),
                 new Promise<void>((resolve) => {
                     this.io!.quit(() => {
                         resolve()