speechflow 2.2.1 → 2.3.0

package/speechflow-cli/src/speechflow-node-a2t-openai.ts

@@ -1,6 +1,6 @@
 /*
 **  SpeechFlow - Speech Processing Flow Graph
-**  Copyright (c) 2024-2025 Dr. Ralf S. Engelschall <rse@engelschall.com>
+**  Copyright (c) 2024-2026 Dr. Ralf S. Engelschall <rse@engelschall.com>
 **  Licensed under GPL 3.0 <https://spdx.org/licenses/GPL-3.0-only>
 */
 
@@ -25,7 +25,7 @@ export default class SpeechFlowNodeA2TOpenAI extends SpeechFlowNode {
     /*  internal state  */
     private openai:            OpenAI                                    | null = null
     private ws:                ws.WebSocket                              | null = null
-    private queue:             util.SingleQueue<SpeechFlowChunk | null>  | null = null
+    private queue:             util.AsyncQueue<SpeechFlowChunk | null>   | null = null
     private resampler:         SpeexResampler                            | null = null
     private closing                                                             = false
     private connectionTimeout: ReturnType<typeof setTimeout>             | null = null
@@ -67,7 +67,7 @@ export default class SpeechFlowNodeA2TOpenAI extends SpeechFlowNode {
         this.closing = false
 
         /*  create queue for results  */
-        this.queue = new util.SingleQueue<SpeechFlowChunk | null>()
+        this.queue = new util.AsyncQueue<SpeechFlowChunk | null>()
 
         /*  create a store for the meta information  */
         const metastore = new util.TimeStore<Map<string, any>>()
@@ -139,10 +139,6 @@ export default class SpeechFlowNodeA2TOpenAI extends SpeechFlowNode {
         })
 
         /*  hook onto session events  */
-        this.ws.on("open", () => {
-            this.log("info", "WebSocket connection opened")
-            sendMessage({ type: "transcription.create" })
-        })
         this.ws.on("close", () => {
             this.log("info", "WebSocket connection closed")
             if (!this.closing && this.queue !== null)
@@ -167,8 +163,11 @@ export default class SpeechFlowNodeA2TOpenAI extends SpeechFlowNode {
             }, new Map<string, any>())
         }
 
-        /*  track transcription text  */
-        let text = ""
+        /*  remember opening time to receive time zero offset  */
+        this.timeOpen = DateTime.now()
+
+        /*  track transcription text per item  */
+        const textByItem = new Map<string, string>()
         this.ws.on("message", (data) => {
             let ev: Record<string, unknown>
             try {
@@ -186,13 +185,16 @@ export default class SpeechFlowNodeA2TOpenAI extends SpeechFlowNode {
                 case "transcription_session.created":
                     break
                 case "conversation.item.created": {
-                    text = ""
+                    const itemId = (ev.item as Record<string, unknown>)?.id as string
+                    if (itemId)
+                        textByItem.set(itemId, "")
                     break
                 }
                 case "conversation.item.input_audio_transcription.delta": {
-                    text += ev.delta as string
+                    const itemId = ev.item_id as string
+                    const text   = (textByItem.get(itemId) ?? "") + (ev.delta as string)
+                    textByItem.set(itemId, text)
                     if (this.params.interim && !this.closing && this.queue !== null) {
-                        const itemId = ev.item_id as string
                         const timing = speechTiming.get(itemId)
                         const start  = timing !== undefined ? Duration.fromMillis(timing.startMs) : DateTime.now().diff(this.timeOpen!)
                         const end    = timing !== undefined ? Duration.fromMillis(timing.endMs)   : start
@@ -204,7 +206,7 @@ export default class SpeechFlowNodeA2TOpenAI extends SpeechFlowNode {
                 }
                 case "conversation.item.input_audio_transcription.completed": {
                     if (!this.closing && this.queue !== null) {
-                        text = ev.transcript as string
+                        const text   = ev.transcript as string
                         const itemId = ev.item_id as string
                         const timing = speechTiming.get(itemId)
                         const start  = timing !== undefined ? Duration.fromMillis(timing.startMs) : DateTime.now().diff(this.timeOpen!)
@@ -213,8 +215,8 @@ export default class SpeechFlowNodeA2TOpenAI extends SpeechFlowNode {
                         chunk.meta = aggregateMeta(start, end)
                         metastore.prune(start)
                         speechTiming.delete(itemId)
+                        textByItem.delete(itemId)
                         this.queue.write(chunk)
-                        text = ""
                     }
                     break
                 }
@@ -248,9 +250,6 @@ export default class SpeechFlowNodeA2TOpenAI extends SpeechFlowNode {
             }
         })
 
-        /*  remember opening time to receive time zero offset  */
-        this.timeOpen = DateTime.now()
-
         /*  provide Duplex stream and internally attach to OpenAI API  */
         const self = this
         const reads = new util.PromiseSet<void>()
@@ -260,7 +259,7 @@ export default class SpeechFlowNodeA2TOpenAI extends SpeechFlowNode {
             decodeStrings:      false,
             highWaterMark:      1,
             write (chunk: SpeechFlowChunk, encoding, callback) {
-                if (self.closing || self.ws === null) {
+                if (self.closing || self.ws === null || self.resampler === null) {
                     callback(new Error("stream already destroyed"))
                     return
                 }
@@ -274,7 +273,7 @@ export default class SpeechFlowNodeA2TOpenAI extends SpeechFlowNode {
                         if (chunk.meta.size > 0)
                             metastore.store(chunk.timestampStart, chunk.timestampEnd, chunk.meta)
                         try {
-                            const payload = self.resampler!.processChunk(chunk.payload)
+                            const payload = self.resampler.processChunk(chunk.payload)
                             const audioB64 = payload.toString("base64")
                             sendMessage({
                                 type: "input_audio_buffer.append",
@@ -296,17 +295,23 @@ export default class SpeechFlowNodeA2TOpenAI extends SpeechFlowNode {
                 }
                 try {
                     sendMessage({ type: "input_audio_buffer.commit" })
-                    self.ws.close()
-                    await util.sleep(50)
+                    self.ws?.close()
+                    await new Promise<void>((resolve) => {
+                        const timeout = setTimeout(() => { resolve() }, 5000)
+                        self.ws?.once("close", () => {
+                            clearTimeout(timeout)
+                            resolve()
+                        })
+                    })
                 }
                 catch (error) {
                     self.log("warning", `error closing OpenAI connection: ${error}`)
                 }
+
+                /*  await all read operations  */
                 await reads.awaitAll()
-                const chunks: Array<SpeechFlowChunk | null> = self.queue?.drain() ?? []
-                for (const chunk of chunks)
-                    this.push(chunk)
-                this.push(null)
+
+                /*  NOTICE: do not push null here -- let the WebSocket close event handle it  */
                 callback()
             },
             read (size) {
@@ -346,6 +351,12 @@ export default class SpeechFlowNodeA2TOpenAI extends SpeechFlowNode {
             this.connectionTimeout = null
         }
 
+        /*  shutdown stream  */
+        if (this.stream !== null) {
+            await util.destroyStream(this.stream)
+            this.stream = null
+        }
+
         /*  signal EOF to any pending read operations  */
         if (this.queue !== null) {
             this.queue.write(null)
@@ -362,12 +373,9 @@ export default class SpeechFlowNodeA2TOpenAI extends SpeechFlowNode {
             this.openai = null
 
         /*  close resampler  */
-        this.resampler = null
-
-        /*  shutdown stream  */
-        if (this.stream !== null) {
-            await util.destroyStream(this.stream)
-            this.stream = null
+        if (this.resampler !== null) {
+            this.resampler.destroy()
+            this.resampler = null
         }
     }
 }

package/speechflow-cli/src/speechflow-node-t2a-amazon.ts

@@ -1,6 +1,6 @@
 /*
 **  SpeechFlow - Speech Processing Flow Graph
-**  Copyright (c) 2024-2025 Dr. Ralf S. Engelschall <rse@engelschall.com>
+**  Copyright (c) 2024-2026 Dr. Ralf S. Engelschall <rse@engelschall.com>
 **  Licensed under GPL 3.0 <https://spdx.org/licenses/GPL-3.0-only>
 */
 
@@ -131,9 +131,13 @@ export default class SpeechFlowNodeT2AAmazon extends SpeechFlowNode {
                 else if (chunk.payload === "")
                     callback()
                 else {
+                    let callbackCalled = false
                     let processTimeout: ReturnType<typeof setTimeout> | null = setTimeout(() => {
                         processTimeout = null
-                        callback(new Error("AWS Polly API timeout"))
+                        if (!callbackCalled) {
+                            callbackCalled = true
+                            callback(new Error("AWS Polly API timeout"))
+                        }
                     }, 60 * 1000)
                     const clearProcessTimeout = () => {
                         if (processTimeout !== null) {
@@ -143,8 +147,11 @@ export default class SpeechFlowNodeT2AAmazon extends SpeechFlowNode {
                     }
                     self.log("debug", `send data (${chunk.payload.length} bytes): "${chunk.payload}"`)
                     textToSpeech(chunk.payload as string).then((buffer) => {
+                        clearProcessTimeout()
+                        if (callbackCalled)
+                            return
+                        callbackCalled = true
                         if (self.closing) {
-                            clearProcessTimeout()
                             callback(new Error("stream destroyed during processing"))
                             return
                         }
@@ -157,11 +164,13 @@ export default class SpeechFlowNodeT2AAmazon extends SpeechFlowNode {
                         chunkNew.type         = "audio"
                         chunkNew.payload      = buffer
                         chunkNew.timestampEnd = Duration.fromMillis(chunkNew.timestampStart.toMillis() + durationMs)
-                        clearProcessTimeout()
                         this.push(chunkNew)
                         callback()
                     }).catch((error: unknown) => {
                         clearProcessTimeout()
+                        if (callbackCalled)
+                            return
+                        callbackCalled = true
                         callback(util.ensureError(error, "AWS Polly processing failed"))
                     })
                 }
@@ -184,8 +193,10 @@ export default class SpeechFlowNodeT2AAmazon extends SpeechFlowNode {
         }
 
         /*  destroy resampler  */
-        if (this.resampler !== null)
+        if (this.resampler !== null) {
+            this.resampler.destroy()
             this.resampler = null
+        }
 
         /*  destroy AWS Polly API  */
         if (this.client !== null) {

package/speechflow-cli/src/speechflow-node-t2a-elevenlabs.ts

@@ -1,6 +1,6 @@
 /*
 **  SpeechFlow - Speech Processing Flow Graph
-**  Copyright (c) 2024-2025 Dr. Ralf S. Engelschall <rse@engelschall.com>
+**  Copyright (c) 2024-2026 Dr. Ralf S. Engelschall <rse@engelschall.com>
 **  Licensed under GPL 3.0 <https://spdx.org/licenses/GPL-3.0-only>
 */
 
@@ -150,9 +150,13 @@ export default class SpeechFlowNodeT2AElevenlabs extends SpeechFlowNode {
                 else if (chunk.payload === "")
                     callback()
                 else {
+                    let callbackCalled = false
                     let processTimeout: ReturnType<typeof setTimeout> | null = setTimeout(() => {
                         processTimeout = null
-                        callback(new Error("ElevenLabs API timeout"))
+                        if (!callbackCalled) {
+                            callbackCalled = true
+                            callback(new Error("ElevenLabs API timeout"))
+                        }
                     }, 60 * 1000)
                     const clearProcessTimeout = () => {
                         if (processTimeout !== null) {
@@ -163,13 +167,17 @@ export default class SpeechFlowNodeT2AElevenlabs extends SpeechFlowNode {
                     try {
                         if (self.closing) {
                             clearProcessTimeout()
+                            callbackCalled = true
                             callback(new Error("stream destroyed during processing"))
                             return
                         }
                         const stream = await speechStream(chunk.payload as string)
                         const buffer = await getStreamAsBuffer(stream)
+                        clearProcessTimeout()
+                        if (callbackCalled)
+                            return
+                        callbackCalled = true
                         if (self.closing) {
-                            clearProcessTimeout()
                             callback(new Error("stream destroyed during processing"))
                             return
                         }
@@ -187,12 +195,14 @@ export default class SpeechFlowNodeT2AElevenlabs extends SpeechFlowNode {
                         chunkNew.type         = "audio"
                         chunkNew.payload      = bufferResampled
                         chunkNew.timestampEnd = Duration.fromMillis(chunkNew.timestampStart.toMillis() + durationMs)
-                        clearProcessTimeout()
                         this.push(chunkNew)
                         callback()
                     }
                     catch (error) {
                         clearProcessTimeout()
+                        if (callbackCalled)
+                            return
+                        callbackCalled = true
                         callback(util.ensureError(error, "ElevenLabs processing failed"))
                     }
                 }
@@ -215,8 +225,10 @@ export default class SpeechFlowNodeT2AElevenlabs extends SpeechFlowNode {
         }
 
         /*  destroy resampler  */
-        if (this.resampler !== null)
+        if (this.resampler !== null) {
+            this.resampler.destroy()
             this.resampler = null
+        }
 
         /*  destroy ElevenLabs API  */
         if (this.elevenlabs !== null)

package/speechflow-cli/src/speechflow-node-t2a-google.ts

@@ -1,6 +1,6 @@
 /*
 **  SpeechFlow - Speech Processing Flow Graph
-**  Copyright (c) 2024-2025 Dr. Ralf S. Engelschall <rse@engelschall.com>
+**  Copyright (c) 2024-2026 Dr. Ralf S. Engelschall <rse@engelschall.com>
 **  Licensed under GPL 3.0 <https://spdx.org/licenses/GPL-3.0-only>
 */
 
@@ -129,9 +129,13 @@ export default class SpeechFlowNodeT2AGoogle extends SpeechFlowNode {
                 else if (chunk.payload === "")
                     callback()
                 else {
+                    let callbackCalled = false
                     let processTimeout: ReturnType<typeof setTimeout> | null = setTimeout(() => {
                         processTimeout = null
-                        callback(new Error("Google TTS API timeout"))
+                        if (!callbackCalled) {
+                            callbackCalled = true
+                            callback(new Error("Google TTS API timeout"))
+                        }
                     }, 60 * 1000)
                     const clearProcessTimeout = () => {
                         if (processTimeout !== null) {
@@ -142,12 +146,16 @@ export default class SpeechFlowNodeT2AGoogle extends SpeechFlowNode {
                     try {
                         if (self.closing) {
                             clearProcessTimeout()
+                            callbackCalled = true
                             callback(new Error("stream destroyed during processing"))
                             return
                         }
                         const buffer = await textToSpeech(chunk.payload as string)
+                        clearProcessTimeout()
+                        if (callbackCalled)
+                            return
+                        callbackCalled = true
                         if (self.closing) {
-                            clearProcessTimeout()
                             callback(new Error("stream destroyed during processing"))
                             return
                         }
@@ -161,12 +169,14 @@ export default class SpeechFlowNodeT2AGoogle extends SpeechFlowNode {
                         chunkNew.type         = "audio"
                         chunkNew.payload      = buffer
                         chunkNew.timestampEnd = Duration.fromMillis(chunkNew.timestampStart.toMillis() + durationMs)
-                        clearProcessTimeout()
                         this.push(chunkNew)
                         callback()
                     }
                     catch (error) {
                         clearProcessTimeout()
+                        if (callbackCalled)
+                            return
+                        callbackCalled = true
                         callback(util.ensureError(error, "Google TTS processing failed"))
                     }
                 }
@@ -189,8 +199,10 @@ export default class SpeechFlowNodeT2AGoogle extends SpeechFlowNode {
         }
 
         /*  destroy resampler  */
-        if (this.resampler !== null)
+        if (this.resampler !== null) {
+            this.resampler.destroy()
             this.resampler = null
+        }
 
         /*  destroy Google TTS client  */
         if (this.client !== null) {

package/speechflow-cli/src/speechflow-node-t2a-kitten.ts

@@ -0,0 +1,178 @@
+/*
+**  SpeechFlow - Speech Processing Flow Graph
+**  Copyright (c) 2024-2026 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"
+
+/*  external dependencies  */
+import { KittenTTS }  from "kitten-tts-js"
+import { Duration }   from "luxon"
+import SpeexResampler from "speex-resampler"
+
+/*  internal dependencies  */
+import SpeechFlowNode, { SpeechFlowChunk } from "./speechflow-node"
+import * as util                           from "./speechflow-util"
+
+/*  SpeechFlow node for Kitten text-to-speech conversion  */
+export default class SpeechFlowNodeT2AKitten extends SpeechFlowNode {
+    /*  declare official node name  */
+    public static name = "t2a-kitten"
+
+    /*  internal state  */
+    private kitten:    KittenTTS      | null = null
+    private resampler: SpeexResampler | null = null
+    private closing                          = false
+
+    /*  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({
+            model:    { type: "string", val: "KittenML/kitten-tts-nano-0.8", pos: 0, match: /^.+$/ },
+            voice:    { type: "string", val: "Bruno",  pos: 1, match: /^(?:Bella|Jasper|Luna|Bruno|Rosie|Hugo|Kiki|Leo)$/ },
+            speed:    { type: "number", val: 1.25,     pos: 2, match: (n: number) => n >= 0.5 && n <= 2.0 }
+        })
+
+        /*  declare node input/output format  */
+        this.input  = "text"
+        this.output = "audio"
+    }
+
+    /*  one-time status of node  */
+    async status () {
+        return {}
+    }
+
+    /*  open node  */
+    async open () {
+        /*  clear destruction flag  */
+        this.closing = false
+
+        /*  establish Kitten TTS  */
+        this.kitten = await KittenTTS.from_pretrained(this.params.model)
+        if (this.kitten === null)
+            throw new Error("failed to instantiate Kitten TTS")
+
+        /*  establish resampler from Kitten's 24Khz
+            output to our standard audio sample rate (48KHz)  */
+        this.resampler = new SpeexResampler(1, 24000, this.config.audioSampleRate, 7)
+
+        /*  perform text-to-speech operation with Kitten TTS API  */
+        const text2speech = async (text: string) => {
+            this.log("info", `Kitten TTS: input: "${text}"`)
+            const audio = await this.kitten!.generate(text, {
+                voice: this.params.voice,
+                speed: this.params.speed
+            })
+            if (audio.sampling_rate !== 24000)
+                throw new Error("expected 24KHz sampling rate in Kitten TTS output")
+
+            /*  convert audio samples from PCM/F32/24Khz to PCM/I16/24KHz  */
+            const samples = audio.data
+            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  */
+            if (this.resampler === null)
+                throw new Error("resampler already destroyed")
+            return this.resampler.processChunk(buffer1)
+        }
+
+        /*  create transform stream and connect it to the Kitten TTS API  */
+        const self = this
+        this.stream = new Stream.Transform({
+            writableObjectMode: true,
+            readableObjectMode: true,
+            decodeStrings:      false,
+            highWaterMark:      1,
+            transform (chunk: SpeechFlowChunk, encoding, callback) {
+                if (self.closing)
+                    callback(new Error("stream already destroyed"))
+                else if (Buffer.isBuffer(chunk.payload))
+                    callback(new Error("invalid chunk payload type"))
+                else if (chunk.payload === "")
+                    callback()
+                else {
+                    let callbackCalled = false
+                    let processTimeout: ReturnType<typeof setTimeout> | null = setTimeout(() => {
+                        processTimeout = null
+                        if (!callbackCalled) {
+                            callbackCalled = true
+                            callback(new Error("Kitten TTS timeout"))
+                        }
+                    }, 60 * 1000)
+                    const clearProcessTimeout = () => {
+                        if (processTimeout !== null) {
+                            clearTimeout(processTimeout)
+                            processTimeout = null
+                        }
+                    }
+                    text2speech(chunk.payload).then((buffer) => {
+                        clearProcessTimeout()
+                        if (callbackCalled)
+                            return
+                        callbackCalled = true
+                        if (self.closing) {
+                            callback(new Error("stream destroyed during processing"))
+                            return
+                        }
+                        self.log("info", `Kitten TTS: received audio (buffer length: ${buffer.byteLength})`)
+
+                        /*  calculate actual audio duration from PCM buffer size  */
+                        const durationMs = util.audioBufferDuration(buffer,
+                            self.config.audioSampleRate, self.config.audioBitDepth) * 1000
+
+                        /*  create new chunk with recalculated timestamps  */
+                        const chunkNew        = chunk.clone()
+                        chunkNew.type         = "audio"
+                        chunkNew.payload      = buffer
+                        chunkNew.timestampEnd = Duration.fromMillis(chunkNew.timestampStart.toMillis() + durationMs)
+                        this.push(chunkNew)
+                        callback()
+                    }).catch((error: unknown) => {
+                        clearProcessTimeout()
+                        if (callbackCalled)
+                            return
+                        callbackCalled = true
+                        callback(util.ensureError(error, "Kitten TTS processing failed"))
+                    })
+                }
+            },
+            final (callback) {
+                callback()
+            }
+        })
+    }
+
+    /*  close node  */
+    async close () {
+        /*  indicate closing  */
+        this.closing = true
+
+        /*  shutdown stream  */
+        if (this.stream !== null) {
+            await util.destroyStream(this.stream)
+            this.stream = null
+        }
+
+        /*  destroy resampler  */
+        if (this.resampler !== null) {
+            this.resampler.destroy()
+            this.resampler = null
+        }
+
+        /*  destroy Kitten TTS API  */
+        if (this.kitten !== null) {
+            await this.kitten.release()
+            this.kitten = null
+        }
+    }
+}
+

package/speechflow-cli/src/speechflow-node-t2a-kokoro.ts

@@ -1,6 +1,6 @@
 /*
 **  SpeechFlow - Speech Processing Flow Graph
-**  Copyright (c) 2024-2025 Dr. Ralf S. Engelschall <rse@engelschall.com>
+**  Copyright (c) 2024-2026 Dr. Ralf S. Engelschall <rse@engelschall.com>
 **  Licensed under GPL 3.0 <https://spdx.org/licenses/GPL-3.0-only>
 */
 
@@ -81,11 +81,12 @@ export default class SpeechFlowNodeT2AKokoro extends SpeechFlowNode {
         this.kokoro = await KokoroTTS.from_pretrained(model, {
             dtype: "q4f16",
             progress_callback: progressCallback
+        }).finally(() => {
+            if (interval !== null) {
+                clearInterval(interval)
+                interval = null
+            }
         })
-        if (interval !== null) {
-            clearInterval(interval)
-            interval = null
-        }
         if (this.kokoro === null)
             throw new Error("failed to instantiate Kokoro")
 
@@ -141,9 +142,13 @@ export default class SpeechFlowNodeT2AKokoro extends SpeechFlowNode {
                 else if (chunk.payload === "")
                     callback()
                 else {
+                    let callbackCalled = false
                     let processTimeout: ReturnType<typeof setTimeout> | null = setTimeout(() => {
                         processTimeout = null
-                        callback(new Error("Kokoro TTS timeout"))
+                        if (!callbackCalled) {
+                            callbackCalled = true
+                            callback(new Error("Kokoro TTS timeout"))
+                        }
                     }, 60 * 1000)
                     const clearProcessTimeout = () => {
                         if (processTimeout !== null) {
@@ -152,8 +157,11 @@ export default class SpeechFlowNodeT2AKokoro extends SpeechFlowNode {
                         }
                     }
                     text2speech(chunk.payload).then((buffer) => {
+                        clearProcessTimeout()
+                        if (callbackCalled)
+                            return
+                        callbackCalled = true
                         if (self.closing) {
-                            clearProcessTimeout()
                             callback(new Error("stream destroyed during processing"))
                             return
                         }
@@ -168,11 +176,13 @@ export default class SpeechFlowNodeT2AKokoro extends SpeechFlowNode {
                         chunkNew.type         = "audio"
                         chunkNew.payload      = buffer
                         chunkNew.timestampEnd = Duration.fromMillis(chunkNew.timestampStart.toMillis() + durationMs)
-                        clearProcessTimeout()
                         this.push(chunkNew)
                         callback()
                     }).catch((error: unknown) => {
                         clearProcessTimeout()
+                        if (callbackCalled)
+                            return
+                        callbackCalled = true
                         callback(util.ensureError(error, "Kokoro processing failed"))
                     })
                 }
@@ -195,8 +205,10 @@ export default class SpeechFlowNodeT2AKokoro extends SpeechFlowNode {
         }
 
         /*  destroy resampler  */
-        if (this.resampler !== null)
+        if (this.resampler !== null) {
+            this.resampler.destroy()
             this.resampler = null
+        }
 
         /*  destroy Kokoro API  */
         if (this.kokoro !== null)