speechflow 1.6.5 → 1.6.6

package/speechflow-cli/src/speechflow-node-a2a-filler.ts

@@ -44,14 +44,14 @@ class AudioFiller extends EventEmitter {
     }
 
     /*  emit a chunk of silence  */
-    private emitSilence (fromSamples: number, toSamples: number) {
+    private emitSilence (fromSamples: number, toSamples: number, meta?: Map<string, any>) {
         const frames = Math.max(0, Math.floor(toSamples - fromSamples))
         if (frames <= 0)
             return
         const payload = Buffer.alloc(frames * this.bytesPerFrame) /* already zeroed */
         const timestampStart = this.durationFromSamples(fromSamples)
         const timestampEnd   = this.durationFromSamples(toSamples)
-        const chunk = new SpeechFlowChunk(timestampStart, timestampEnd, "final", "audio", payload)
+        const chunk = new SpeechFlowChunk(timestampStart, timestampEnd, "final", "audio", payload, meta ? new Map(meta) : undefined)
         this.emit("chunk", chunk)
     }
 
@@ -64,7 +64,7 @@ class AudioFiller extends EventEmitter {
 
         /*  if chunk starts beyond what we've emitted, insert silence for the gap  */
         if (startSamp > this.emittedEndSamples + this.sampleTolerance) {
-            this.emitSilence(this.emittedEndSamples, startSamp)
+            this.emitSilence(this.emittedEndSamples, startSamp, chunk.meta)
             this.emittedEndSamples = startSamp
         }
 
@@ -95,7 +95,7 @@ class AudioFiller extends EventEmitter {
         const outEndSamples   = outStartSamples + Math.floor(payload.length / this.bytesPerFrame)
         const timestampStart  = this.durationFromSamples(outStartSamples)
         const timestampEnd    = this.durationFromSamples(outEndSamples)
-        const c = new SpeechFlowChunk(timestampStart, timestampEnd, "final", "audio", payload)
+        const c = new SpeechFlowChunk(timestampStart, timestampEnd, "final", "audio", payload, new Map(chunk.meta))
         this.emit("chunk", c)
 
         /*  advance emitted cursor  */
@@ -109,7 +109,7 @@ export default class SpeechFlowNodeA2AFiller extends SpeechFlowNode {
     public static name = "a2a-filler"
 
     /*  internal state  */
-    private destroyed = false
+    private closing = false
     private filler: AudioFiller | null = null
     private sendQueue: util.AsyncQueue<SpeechFlowChunk | null> | null = null
 
@@ -130,7 +130,7 @@ export default class SpeechFlowNodeA2AFiller extends SpeechFlowNode {
     /*  open node  */
     async open () {
         /*  clear destruction flag  */
-        this.destroyed = false
+        this.closing = false
 
         /*  establish queues  */
         this.filler  = new AudioFiller(this.config.audioSampleRate, this.config.audioChannels)
@@ -148,27 +148,29 @@ export default class SpeechFlowNodeA2AFiller extends SpeechFlowNode {
             writableObjectMode: true,
             decodeStrings:      false,
             write (chunk: SpeechFlowChunk & { type: "audio", payload: Buffer }, encoding, callback) {
-                if (self.destroyed || self.filler === null)
+                if (self.closing || self.filler === null)
                     callback(new Error("stream already destroyed"))
                 else if (!Buffer.isBuffer(chunk.payload))
                     callback(new Error("invalid chunk payload type"))
                 else {
                     try {
+                        if (self.closing || self.filler === null)
+                            throw new Error("stream already destroyed")
                         self.filler.add(chunk)
                         callback()
                     }
-                    catch (error: any) {
-                        callback(error)
+                    catch (error: unknown) {
+                        callback(util.ensureError(error))
                     }
                 }
             },
             read (size) {
-                if (self.destroyed || self.sendQueue === null) {
+                if (self.closing || self.sendQueue === null) {
                     this.push(null)
                     return
                 }
                 self.sendQueue.read().then((chunk) => {
-                    if (self.destroyed) {
+                    if (self.closing || self.sendQueue === null) {
                         this.push(null)
                         return
                     }
@@ -181,12 +183,12 @@ export default class SpeechFlowNodeA2AFiller extends SpeechFlowNode {
                         this.push(chunk)
                     }
                 }).catch((error: unknown) => {
-                    if (!self.destroyed)
+                    if (!self.closing && self.sendQueue !== null)
                         self.log("error", `queue read error: ${util.ensureError(error).message}`)
                 })
             },
             final (callback) {
-                if (self.destroyed) {
+                if (self.closing) {
                     callback()
                     return
                 }
@@ -198,8 +200,8 @@ export default class SpeechFlowNodeA2AFiller extends SpeechFlowNode {
 
     /*  close node  */
     async close () {
-        /*  indicate destruction  */
-        this.destroyed = true
+        /*  indicate closing  */
+        this.closing = true
 
         /*  destroy queues  */
         if (this.sendQueue !== null) {
@@ -213,9 +215,9 @@ export default class SpeechFlowNodeA2AFiller extends SpeechFlowNode {
             this.filler = null
         }
 
-        /*  close stream  */
+        /*  shutdown stream  */
         if (this.stream !== null) {
-            this.stream.destroy()
+            await util.destroyStream(this.stream)
             this.stream = null
         }
     }

package/speechflow-cli/src/speechflow-node-a2a-gain.ts

@@ -17,7 +17,7 @@ export default class SpeechFlowNodeA2AGain extends SpeechFlowNode {
     public static name = "a2a-gain"
 
     /*  internal state  */
-    private destroyed = false
+    private closing = false
 
     /*  construct node  */
     constructor (id: string, cfg: { [ id: string ]: any }, opts: { [ id: string ]: any }, args: any[]) {
@@ -36,7 +36,7 @@ export default class SpeechFlowNodeA2AGain extends SpeechFlowNode {
     /*  open node  */
     async open () {
         /*  clear destruction flag  */
-        this.destroyed = false
+        this.closing = false
 
         /*  adjust gain  */
         const adjustGain = (chunk: SpeechFlowChunk & { payload: Buffer }, db: number) => {
@@ -57,7 +57,7 @@ export default class SpeechFlowNodeA2AGain extends SpeechFlowNode {
             writableObjectMode: true,
             decodeStrings:      false,
             transform (chunk: SpeechFlowChunk & { payload: Buffer }, encoding, callback) {
-                if (self.destroyed) {
+                if (self.closing) {
                     callback(new Error("stream already destroyed"))
                     return
                 }
@@ -73,7 +73,7 @@ export default class SpeechFlowNodeA2AGain extends SpeechFlowNode {
                 }
             },
             final (callback) {
-                if (self.destroyed) {
+                if (self.closing) {
                     callback()
                     return
                 }
@@ -85,12 +85,12 @@ export default class SpeechFlowNodeA2AGain extends SpeechFlowNode {
 
     /*  close node  */
     async close () {
-        /*  indicate destruction  */
-        this.destroyed = true
+        /*  indicate closing  */
+        this.closing = true
 
-        /*  close stream  */
+        /*  shutdown stream  */
         if (this.stream !== null) {
-            this.stream.destroy()
+            await util.destroyStream(this.stream)
             this.stream = null
         }
     }

package/speechflow-cli/src/speechflow-node-a2a-gender.ts

@@ -12,17 +12,19 @@ import Stream  from "node:stream"
 import * as Transformers     from "@huggingface/transformers"
 import { WaveFile }          from "wavefile"
 import { getRMS, AudioData } from "audio-inspect"
+import { Duration }          from "luxon"
 
 /*  internal dependencies  */
 import SpeechFlowNode, { SpeechFlowChunk } from "./speechflow-node"
 import * as util                           from "./speechflow-util"
 
 /*  audio stream queue element */
+type Gender = "male" | "female" | "unknown"
 type AudioQueueElement = {
     type:         "audio-frame",
     chunk:        SpeechFlowChunk,
     data:         Float32Array,
-    gender?:      "male" | "female" | "unknown"
+    gender?:      Gender
 } | {
     type:         "audio-eof"
 }
@@ -38,7 +40,7 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
     private queueRecv = this.queue.pointerUse("recv")
     private queueAC   = this.queue.pointerUse("ac")
     private queueSend = this.queue.pointerUse("send")
-    private shutdown  = false
+    private closing = false
     private workingOffTimer:  ReturnType<typeof setTimeout>  | null = null
     private progressInterval: ReturnType<typeof setInterval> | null = null
 
@@ -65,8 +67,8 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
         if (this.config.audioBitDepth !== 16 || !this.config.audioLittleEndian)
             throw new Error("Gender node currently supports PCM-S16LE audio only")
 
-        /*  clear shutdown flag  */
-        this.shutdown = false
+        /*  clear destruction flag  */
+        this.closing = false
 
         /*  the used model  */
         const model = "Xenova/wav2vec2-large-xlsr-53-gender-recognition-librispeech"
@@ -74,7 +76,7 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
         /*  track download progress when instantiating Transformers engine and model  */
         const progressState = new Map<string, number>()
         const progressCallback: Transformers.ProgressCallback = (progress: any) => {
-            if (this.shutdown)
+            if (this.closing)
                 return
             let artifact = model
             if (typeof progress.file === "string")
@@ -88,7 +90,7 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
                 progressState.set(artifact, percent)
         }
         this.progressInterval = setInterval(() => {
-            if (this.shutdown)
+            if (this.closing)
                 return
             for (const [ artifact, percent ] of progressState) {
                 this.log("info", `downloaded ${percent.toFixed(2)}% of artifact "${artifact}"`)
@@ -126,9 +128,10 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
         const sampleRateTarget = 16000
 
         /*  classify a single large-enough concatenated audio frame  */
+        let genderLast: Gender = "unknown"
         const classify = async (data: Float32Array) => {
-            if (this.shutdown || this.classifier === null)
-                throw new Error("classifier shutdown during operation")
+            if (this.closing || this.classifier === null)
+                throw new Error("classifier destroyed during operation")
 
             /*  check volume level and return "unknown" if too low
                 in order to avoid a wrong classificaton  */
@@ -141,7 +144,7 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
             } satisfies AudioData
             const rms = getRMS(audioData, { asDB: true })
             if (rms < this.params.volumeThreshold)
-                return "unknown"
+                return genderLast
 
             /*  classify audio  */
             const result = await Promise.race([
@@ -157,22 +160,25 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
             const female = c2 ? c2.score : 0.0
             const threshold  = this.params.threshold
             const hysteresis = this.params.hysteresis
+            let genderNow: Gender = genderLast
             if (male > threshold && male > female + hysteresis)
-                return "male"
+                genderNow = "male"
             else if (female > threshold && female > male + hysteresis)
-                return "female"
-            else
-                return "unknown"
+                genderNow = "female"
+            if (genderNow !== genderLast) {
+                this.log("info", `switching detected gender from <${genderLast}> to <${genderNow}>`)
+                genderLast = genderNow
+            }
+            return genderNow
         }
 
         /*  work off queued audio frames  */
         const frameWindowDuration = this.params.window / 1000
         const frameWindowSamples  = Math.floor(frameWindowDuration * sampleRateTarget)
-        let lastGender = ""
         let workingOff = false
         const workOffQueue = async () => {
             /*  control working off round  */
-            if (workingOff || this.shutdown)
+            if (workingOff || this.closing)
                 return
             workingOff = true
             if (this.workingOffTimer !== null) {
@@ -189,7 +195,7 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
                 data.fill(0)
                 let samples = 0
                 let pos = pos0
-                while (pos < posL && samples < frameWindowSamples && !this.shutdown) {
+                while (pos < posL && samples < frameWindowSamples && !this.closing) {
                     const element = this.queueAC.peek(pos)
                     if (element === undefined || element.type !== "audio-frame")
                         break
@@ -199,12 +205,12 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
                     }
                     pos++
                 }
-                if (pos0 < pos && samples > frameWindowSamples * 0.75 && !this.shutdown) {
+                if (pos0 < pos && samples > frameWindowSamples * 0.75 && !this.closing) {
                     const gender = await classify(data)
-                    if (this.shutdown)
+                    if (this.closing)
                         return
                     const posM = pos0 + Math.trunc((pos - pos0) * 0.25)
-                    while (pos0 < posM && pos0 < posL && !this.shutdown) {
+                    while (pos0 < posM && pos0 < posL && !this.closing) {
                         const element = this.queueAC.peek(pos0)
                         if (element === undefined || element.type !== "audio-frame")
                             break
@@ -213,10 +219,6 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
                         this.queueAC.walk(+1)
                         pos0++
                     }
-                    if (lastGender !== gender && !this.shutdown) {
-                        this.log("info", `gender now recognized as <${gender}>`)
-                        lastGender = gender
-                    }
                 }
             }
             catch (error) {
@@ -225,7 +227,7 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
 
             /*  re-initiate working off round  */
             workingOff = false
-            if (!this.shutdown) {
+            if (!this.closing) {
                 this.workingOffTimer = setTimeout(workOffQueue, 100)
                 this.queue.once("write", workOffQueue)
             }
@@ -242,7 +244,7 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
 
             /*  receive audio chunk (writable side of stream)  */
             write (chunk: SpeechFlowChunk, encoding, callback) {
-                if (self.shutdown) {
+                if (self.closing) {
                     callback(new Error("stream already destroyed"))
                     return
                 }
@@ -257,21 +259,21 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
                         const wav = new WaveFile()
                         wav.fromScratch(self.config.audioChannels, self.config.audioSampleRate, "32f", data)
                         wav.toSampleRate(sampleRateTarget, { method: "cubic" })
-                        data = wav.getSamples(false, Float32Array) as any as Float32Array<ArrayBuffer>
+                        data = wav.getSamples(false, Float32Array) as unknown as Float32Array<ArrayBuffer>
 
                         /*  queue chunk and converted data  */
                         self.queueRecv.append({ type: "audio-frame", chunk, data })
                         callback()
                     }
                     catch (error) {
-                        callback(error instanceof Error ? error : new Error("audio processing failed"))
+                        callback(util.ensureError(error, "audio processing failed"))
                     }
                 }
             },
 
             /*  receive no more audio chunks (writable side of stream)  */
             final (callback) {
-                if (self.shutdown) {
+                if (self.closing) {
                     callback()
                     return
                 }
@@ -285,7 +287,7 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
             read (_size) {
                 /*  flush pending audio chunks  */
                 const flushPendingChunks = () => {
-                    if (self.shutdown) {
+                    if (self.closing) {
                         this.push(null)
                         return
                     }
@@ -297,7 +299,7 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
                         && element.type === "audio-frame"
                         && element.gender !== undefined) {
                         while (true) {
-                            if (self.shutdown) {
+                            if (self.closing) {
                                 this.push(null)
                                 return
                             }
@@ -312,14 +314,18 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
                                 && element.gender === undefined)
                                 break
                             const duration = util.audioArrayDuration(element.data)
-                            self.log("debug", `send chunk (${duration.toFixed(3)}s) with gender <${element.gender}>`)
+                            const fmtTime = (t: Duration) => t.toFormat("hh:mm:ss.SSS")
+                            const times = `start: ${fmtTime(element.chunk.timestampStart)}, ` +
+                                `end: ${fmtTime(element.chunk.timestampEnd)}`
+                            self.log("debug", `send chunk (${times}, duration: ${duration.toFixed(3)}s) ` +
+                                `with gender <${element.gender}>`)
                             element.chunk.meta.set("gender", element.gender)
                             this.push(element.chunk)
                             self.queueSend.walk(+1)
                             self.queue.trim()
                         }
                     }
-                    else if (!self.shutdown)
+                    else if (!self.closing)
                         self.queue.once("write", flushPendingChunks)
                 }
                 flushPendingChunks()
@@ -329,8 +335,8 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
 
     /*  close node  */
     async close () {
-        /*  indicate shutdown  */
-        this.shutdown = true
+        /*  indicate closing  */
+        this.closing = true
 
         /*  cleanup working-off timer  */
         if (this.workingOffTimer !== null) {
@@ -347,9 +353,9 @@ export default class SpeechFlowNodeA2AGender extends SpeechFlowNode {
         /*  remove all event listeners  */
         this.queue.removeAllListeners("write")
 
-        /*  close stream  */
+        /*  shutdown stream  */
         if (this.stream !== null) {
-            this.stream.destroy()
+            await util.destroyStream(this.stream)
             this.stream = null
         }
 

package/speechflow-cli/src/speechflow-node-a2a-meter.ts

@@ -24,7 +24,7 @@ export default class SpeechFlowNodeA2AMeter extends SpeechFlowNode {
     private calcInterval: ReturnType<typeof setInterval> | null = null
     private silenceTimer: ReturnType<typeof setTimeout>  | null = null
     private chunkBuffer = new Float32Array(0)
-    private destroyed = false
+    private closing = false
 
     /*  construct node  */
     constructor (id: string, cfg: { [ id: string ]: any }, opts: { [ id: string ]: any }, args: any[]) {
@@ -52,7 +52,7 @@ export default class SpeechFlowNodeA2AMeter extends SpeechFlowNode {
             throw new Error("meter node currently supports PCM-S16LE audio only")
 
         /*  clear destruction flag  */
-        this.destroyed = false
+        this.closing = false
 
         /*  internal state  */
         let lufsm = -60
@@ -72,7 +72,7 @@ export default class SpeechFlowNodeA2AMeter extends SpeechFlowNode {
         /*  setup chunking interval  */
         this.calcInterval = setInterval(() => {
             /*  short-circuit during destruction  */
-            if (this.destroyed)
+            if (this.closing)
                 return
 
             /*  short-circuit if still not enough chunk data  */
@@ -133,7 +133,7 @@ export default class SpeechFlowNodeA2AMeter extends SpeechFlowNode {
 
         /*  setup loudness emitting interval  */
         this.emitInterval = setInterval(() => {
-            if (this.destroyed)
+            if (this.closing)
                 return
             this.log("debug", `LUFS-M: ${lufsm.toFixed(1)} dB, RMS: ${rms.toFixed(1)} dB`)
             this.sendResponse([ "meter", "LUFS-M", lufsm ])
@@ -152,7 +152,7 @@ export default class SpeechFlowNodeA2AMeter extends SpeechFlowNode {
 
             /*  transform audio chunk  */
             transform (chunk: SpeechFlowChunk, encoding, callback) {
-                if (self.destroyed) {
+                if (self.closing) {
                     callback(new Error("stream already destroyed"))
                     return
                 }
@@ -178,12 +178,12 @@ export default class SpeechFlowNodeA2AMeter extends SpeechFlowNode {
                         callback()
                     }
                     catch (error) {
-                        callback(error instanceof Error ? error : new Error("meter processing failed"))
+                        callback(util.ensureError(error, "meter processing failed"))
                     }
                 }
             },
             final (callback) {
-                if (self.destroyed || self.params.mode === "sink") {
+                if (self.closing || self.params.mode === "sink") {
                     callback()
                     return
                 }
@@ -195,8 +195,8 @@ export default class SpeechFlowNodeA2AMeter extends SpeechFlowNode {
 
     /*  close node  */
     async close () {
-        /*  indicate destruction immediately to stop any ongoing operations  */
-        this.destroyed = true
+        /*  indicate closing immediately to stop any ongoing operations  */
+        this.closing = true
 
         /*  stop intervals  */
         if (this.emitInterval !== null) {
@@ -212,9 +212,9 @@ export default class SpeechFlowNodeA2AMeter extends SpeechFlowNode {
             this.silenceTimer = null
         }
 
-        /*  close stream  */
+        /*  shutdown stream  */
         if (this.stream !== null) {
-            this.stream.destroy()
+            await util.destroyStream(this.stream)
             this.stream = null
         }
     }

package/speechflow-cli/src/speechflow-node-a2a-mute.ts

@@ -9,6 +9,7 @@ import Stream from "node:stream"
 
 /*  internal dependencies  */
 import SpeechFlowNode, { SpeechFlowChunk } from "./speechflow-node"
+import * as util                           from "./speechflow-util"
 
 /*  the type of muting  */
 type MuteMode =
@@ -23,7 +24,7 @@ export default class SpeechFlowNodeA2AMute extends SpeechFlowNode {
 
     /*  internal state  */
     private muteMode: MuteMode = "none"
-    private destroyed = false
+    private closing = false
 
     /*  construct node  */
     constructor (id: string, cfg: { [ id: string ]: any }, opts: { [ id: string ]: any }, args: any[]) {
@@ -39,7 +40,7 @@ export default class SpeechFlowNodeA2AMute extends SpeechFlowNode {
 
     /*  receive external request  */
     async receiveRequest (params: any[]) {
-        if (this.destroyed)
+        if (this.closing)
             throw new Error("mute: node already destroyed")
         try {
             if (params.length === 2 && params[0] === "mode") {
@@ -61,7 +62,7 @@ export default class SpeechFlowNodeA2AMute extends SpeechFlowNode {
 
     /*  change mute mode  */
     setMuteMode (mode: MuteMode) {
-        if (this.destroyed) {
+        if (this.closing) {
             this.log("warning", "attempted to set mute mode on destroyed node")
             return
         }
@@ -72,7 +73,7 @@ export default class SpeechFlowNodeA2AMute extends SpeechFlowNode {
     /*  open node  */
     async open () {
         /*  clear destruction flag  */
-        this.destroyed = false
+        this.closing = false
 
         /*  establish a transform stream  */
         const self = this
@@ -81,7 +82,7 @@ export default class SpeechFlowNodeA2AMute extends SpeechFlowNode {
             writableObjectMode: true,
             decodeStrings:      false,
             transform (chunk: SpeechFlowChunk, encoding, callback) {
-                if (self.destroyed) {
+                if (self.closing) {
                     callback(new Error("stream already destroyed"))
                     return
                 }
@@ -106,7 +107,7 @@ export default class SpeechFlowNodeA2AMute extends SpeechFlowNode {
                 }
             },
             final (callback) {
-                if (self.destroyed) {
+                if (self.closing) {
                     callback()
                     return
                 }
@@ -118,12 +119,12 @@ export default class SpeechFlowNodeA2AMute extends SpeechFlowNode {
 
     /*  close node  */
     async close () {
-        /*  indicate destruction  */
-        this.destroyed = true
+        /*  indicate closing  */
+        this.closing = true
 
-        /*  close stream  */
+        /*  shutdown stream  */
         if (this.stream !== null) {
-            this.stream.destroy()
+            await util.destroyStream(this.stream)
             this.stream = null
         }
     }