@qvac/sdk 0.9.2 → 0.10.1

package/dist/server/bare/plugins/onnx-tts/ops/text-to-speech-stream.js

@@ -0,0 +1,116 @@
+import { getModel } from "../../../registry/model-registry.js";
+import { textToSpeechStreamRequestSchema, } from "../../../../../schemas/index.js";
+import { nowMs } from "../../../../../profiling/index.js";
+import { buildStreamResult, hasDefinedValues } from "../../../../../profiling/model-execution.js";
+import { TextToSpeechStreamFailedError } from "../../../../../utils/errors-server.js";
+import { collectTtsStats, } from "../../../utils/tts-stats.js";
+function hasRunStreaming(model) {
+    return (typeof model === "object" &&
+        model !== null &&
+        "runStreaming" in model &&
+        typeof model.runStreaming === "function");
+}
+// Callers (e.g. LLM token deltas) may write raw Buffer chunks whose
+// boundaries split a multi-byte UTF-8 codepoint. A naive `buf.toString("utf8")`
+// per chunk emits U+FFFD at every split point and silently corrupts CJK,
+// emoji, accented scripts, etc. before they reach the tokenizer. Keep a
+// tail-buffer of at most 3 bytes spanning an incomplete trailing codepoint
+// and decode only the complete-codepoint prefix of each combined chunk.
+function findLastCompleteUtf8End(buf) {
+    const len = buf.length;
+    // A UTF-8 codepoint is at most 4 bytes, so a split can only dangle in
+    // the last 3 bytes. Scan backwards to find the most recent leading byte.
+    for (let i = len - 1; i >= 0 && i >= len - 3; i--) {
+        const b = buf[i];
+        if ((b & 0x80) === 0) {
+            // ASCII byte: everything up to and including i is complete.
+            return i + 1;
+        }
+        if ((b & 0xc0) === 0xc0) {
+            // Leading byte of a multi-byte sequence.
+            let expected;
+            if ((b & 0xe0) === 0xc0)
+                expected = 2;
+            else if ((b & 0xf0) === 0xe0)
+                expected = 3;
+            else if ((b & 0xf8) === 0xf0)
+                expected = 4;
+            else
+                return len; // Invalid leader; let toString emit U+FFFD.
+            return i + expected <= len ? len : i;
+        }
+    }
+    return len;
+}
+async function* buffersToUtf8Fragments(inputStream) {
+    let pending = Buffer.alloc(0);
+    for await (const buf of inputStream) {
+        const combined = pending.length === 0 ? buf : Buffer.concat([pending, buf]);
+        const completeEnd = findLastCompleteUtf8End(combined);
+        if (completeEnd > 0) {
+            const s = combined.subarray(0, completeEnd).toString("utf8");
+            if (s.length > 0) {
+                yield s;
+            }
+        }
+        pending =
+            completeEnd < combined.length
+                ? Buffer.from(combined.subarray(completeEnd))
+                : Buffer.alloc(0);
+    }
+    if (pending.length > 0) {
+        // Stream ended with a dangling partial codepoint — flush as-is so the
+        // tokenizer sees the replacement character rather than silently dropping
+        // the final byte(s).
+        const s = pending.toString("utf8");
+        if (s.length > 0) {
+            yield s;
+        }
+    }
+}
+function buildRunStreamingOptions(request) {
+    const o = {};
+    if (request.accumulateSentences !== undefined) {
+        o["accumulateSentences"] = request.accumulateSentences;
+    }
+    if (request.sentenceDelimiterPreset !== undefined) {
+        o["sentenceDelimiterPreset"] = request.sentenceDelimiterPreset;
+    }
+    if (request.maxBufferScalars !== undefined) {
+        o["maxBufferScalars"] = request.maxBufferScalars;
+    }
+    if (request.flushAfterMs !== undefined) {
+        o["flushAfterMs"] = request.flushAfterMs;
+    }
+    return o;
+}
+export async function* textToSpeechStream(params, inputStream) {
+    const request = textToSpeechStreamRequestSchema.parse(params);
+    const model = getModel(request.modelId);
+    const modelStart = nowMs();
+    if (!hasRunStreaming(model)) {
+        throw new TextToSpeechStreamFailedError("textToSpeechStream requires an ONNX TTS model with runStreaming");
+    }
+    const textSource = buffersToUtf8Fragments(inputStream);
+    const streamOpts = buildRunStreamingOptions(request);
+    const response = await model.runStreaming(textSource, Object.keys(streamOpts).length > 0 ? streamOpts : undefined);
+    for await (const data of response.iterate()) {
+        if (data.outputArray == null) {
+            continue;
+        }
+        const buf = Array.from(data.outputArray);
+        if (buf.length === 0) {
+            continue;
+        }
+        yield {
+            buffer: buf,
+            ...(data.chunkIndex !== undefined ? { chunkIndex: data.chunkIndex } : {}),
+            ...(typeof data.sentenceChunk === "string" && data.sentenceChunk.length > 0
+                ? { sentenceChunk: data.sentenceChunk }
+                : {}),
+        };
+    }
+    const modelExecutionMs = nowMs() - modelStart;
+    const stats = collectTtsStats(response);
+    return buildStreamResult(modelExecutionMs, hasDefinedValues(stats) ? stats : undefined);
+}

package/dist/server/bare/plugins/onnx-tts/ops/text-to-speech.d.ts

@@ -1,7 +1,6 @@
 import { type TtsRequest, type TtsStats } from "../../../../../schemas/index";
-export declare function textToSpeech(params: TtsRequest): AsyncGenerator<{
-    buffer: number[];
-}, {
+import { type TtsOpYield } from "../../../utils/tts-stats";
+export declare function textToSpeech(params: TtsRequest): AsyncGenerator<TtsOpYield, {
     modelExecutionMs: number;
     stats?: TtsStats;
 }>;

package/dist/server/bare/plugins/onnx-tts/ops/text-to-speech.js

@@ -1,34 +1,96 @@
 import { getModel } from "../../../registry/model-registry.js";
-import { ttsRequestSchema } from "../../../../../schemas/index.js";
+import { ttsRequestSchema, } from "../../../../../schemas/index.js";
 import { nowMs } from "../../../../../profiling/index.js";
 import { buildStreamResult, hasDefinedValues } from "../../../../../profiling/model-execution.js";
+import { TextToSpeechFailedError } from "../../../../../utils/errors-server.js";
+import { collectTtsStats, } from "../../../utils/tts-stats.js";
+function hasRunStream(model) {
+    return (typeof model === "object" &&
+        model !== null &&
+        "runStream" in model &&
+        typeof model.runStream === "function");
+}
 export async function* textToSpeech(params) {
-    const { modelId, inputType, text, stream } = ttsRequestSchema.parse(params);
+    const { modelId, inputType, text, stream, sentenceStream, sentenceStreamLocale, sentenceStreamMaxChunkScalars, } = ttsRequestSchema.parse(params);
     const model = getModel(modelId);
     const modelStart = nowMs();
-    const response = (await model.run({ input: text, inputType }));
+    if (sentenceStream) {
+        if (!hasRunStream(model)) {
+            throw new TextToSpeechFailedError("sentenceStream requires an ONNX TTS model with runStream (Supertonic)");
+        }
+        const streamOpts = sentenceStreamLocale !== undefined || sentenceStreamMaxChunkScalars !== undefined
+            ? {
+                ...(sentenceStreamLocale !== undefined
+                    ? { locale: sentenceStreamLocale }
+                    : {}),
+                ...(sentenceStreamMaxChunkScalars !== undefined
+                    ? { maxChunkScalars: sentenceStreamMaxChunkScalars }
+                    : {}),
+            }
+            : undefined;
+        const response = await model.runStream(text, streamOpts);
+        if (!stream) {
+            let completeBuffer = [];
+            for await (const data of response.iterate()) {
+                if (data.outputArray != null) {
+                    completeBuffer = completeBuffer.concat(Array.from(data.outputArray));
+                }
+            }
+            const modelExecutionMs = nowMs() - modelStart;
+            const stats = collectTtsStats(response);
+            yield { buffer: completeBuffer };
+            return buildStreamResult(modelExecutionMs, hasDefinedValues(stats) ? stats : undefined);
+        }
+        for await (const data of response.iterate()) {
+            if (data.outputArray == null)
+                continue;
+            const buf = Array.from(data.outputArray);
+            if (buf.length === 0)
+                continue;
+            yield {
+                buffer: buf,
+                ...(data.chunkIndex !== undefined ? { chunkIndex: data.chunkIndex } : {}),
+                ...(typeof data.sentenceChunk === "string" && data.sentenceChunk.length > 0
+                    ? { sentenceChunk: data.sentenceChunk }
+                    : {}),
+            };
+        }
+        const modelExecutionMs = nowMs() - modelStart;
+        const stats = collectTtsStats(response);
+        return buildStreamResult(modelExecutionMs, hasDefinedValues(stats) ? stats : undefined);
+    }
+    const response = (await model.run({
+        input: text,
+        inputType,
+    }));
     if (!stream) {
-        // Non-streaming mode: collect all chunks and return once
         let completeBuffer = [];
         for await (const data of response.iterate()) {
             completeBuffer = completeBuffer.concat(Array.from(data.outputArray));
         }
         const modelExecutionMs = nowMs() - modelStart;
         const stats = {
-            ...(response.stats?.audioDurationMs !== undefined && { audioDuration: response.stats.audioDurationMs }),
-            ...(response.stats?.totalSamples !== undefined && { totalSamples: response.stats.totalSamples }),
+            ...(response.stats?.audioDurationMs !== undefined && {
+                audioDuration: response.stats.audioDurationMs,
+            }),
+            ...(response.stats?.totalSamples !== undefined && {
+                totalSamples: response.stats.totalSamples,
+            }),
         };
         yield { buffer: completeBuffer };
         return buildStreamResult(modelExecutionMs, hasDefinedValues(stats) ? stats : undefined);
     }
-    // Streaming mode: yield chunks as they arrive
     for await (const data of response.iterate()) {
         yield { buffer: Array.from(data.outputArray) };
     }
     const modelExecutionMs = nowMs() - modelStart;
     const stats = {
-        ...(response.stats?.audioDurationMs !== undefined && { audioDuration: response.stats.audioDurationMs }),
-        ...(response.stats?.totalSamples !== undefined && { totalSamples: response.stats.totalSamples }),
+        ...(response.stats?.audioDurationMs !== undefined && {
+            audioDuration: response.stats.audioDurationMs,
+        }),
+        ...(response.stats?.totalSamples !== undefined && {
+            totalSamples: response.stats.totalSamples,
+        }),
     };
     return buildStreamResult(modelExecutionMs, hasDefinedValues(stats) ? stats : undefined);
 }

package/dist/server/bare/plugins/onnx-tts/plugin.d.ts

@@ -23,6 +23,7 @@ export declare const ttsPlugin: {
             expectedSize: import("zod").ZodOptional<import("zod").ZodNumber>;
             sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
+                embeddings: "embeddings";
                 "llamacpp-completion": "llamacpp-completion";
                 "whispercpp-transcription": "whispercpp-transcription";
                 "llamacpp-embedding": "llamacpp-embedding";
@@ -33,7 +34,6 @@ export declare const ttsPlugin: {
                 "sdcpp-generation": "sdcpp-generation";
                 llm: "llm";
                 whisper: "whisper";
-                embeddings: "embeddings";
                 nmt: "nmt";
                 parakeet: "parakeet";
                 tts: "tts";
@@ -53,6 +53,7 @@ export declare const ttsPlugin: {
             expectedSize: import("zod").ZodOptional<import("zod").ZodNumber>;
             sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
+                embeddings: "embeddings";
                 "llamacpp-completion": "llamacpp-completion";
                 "whispercpp-transcription": "whispercpp-transcription";
                 "llamacpp-embedding": "llamacpp-embedding";
@@ -63,7 +64,6 @@ export declare const ttsPlugin: {
                 "sdcpp-generation": "sdcpp-generation";
                 llm: "llm";
                 whisper: "whisper";
-                embeddings: "embeddings";
                 nmt: "nmt";
                 parakeet: "parakeet";
                 tts: "tts";
@@ -83,6 +83,7 @@ export declare const ttsPlugin: {
             expectedSize: import("zod").ZodOptional<import("zod").ZodNumber>;
             sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
+                embeddings: "embeddings";
                 "llamacpp-completion": "llamacpp-completion";
                 "whispercpp-transcription": "whispercpp-transcription";
                 "llamacpp-embedding": "llamacpp-embedding";
@@ -93,7 +94,6 @@ export declare const ttsPlugin: {
                 "sdcpp-generation": "sdcpp-generation";
                 llm: "llm";
                 whisper: "whisper";
-                embeddings: "embeddings";
                 nmt: "nmt";
                 parakeet: "parakeet";
                 tts: "tts";
@@ -113,6 +113,7 @@ export declare const ttsPlugin: {
             expectedSize: import("zod").ZodOptional<import("zod").ZodNumber>;
             sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
+                embeddings: "embeddings";
                 "llamacpp-completion": "llamacpp-completion";
                 "whispercpp-transcription": "whispercpp-transcription";
                 "llamacpp-embedding": "llamacpp-embedding";
@@ -123,7 +124,6 @@ export declare const ttsPlugin: {
                 "sdcpp-generation": "sdcpp-generation";
                 llm: "llm";
                 whisper: "whisper";
-                embeddings: "embeddings";
                 nmt: "nmt";
                 parakeet: "parakeet";
                 tts: "tts";
@@ -143,6 +143,7 @@ export declare const ttsPlugin: {
             expectedSize: import("zod").ZodOptional<import("zod").ZodNumber>;
             sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
+                embeddings: "embeddings";
                 "llamacpp-completion": "llamacpp-completion";
                 "whispercpp-transcription": "whispercpp-transcription";
                 "llamacpp-embedding": "llamacpp-embedding";
@@ -153,7 +154,6 @@ export declare const ttsPlugin: {
                 "sdcpp-generation": "sdcpp-generation";
                 llm: "llm";
                 whisper: "whisper";
-                embeddings: "embeddings";
                 nmt: "nmt";
                 parakeet: "parakeet";
                 tts: "tts";
@@ -173,6 +173,7 @@ export declare const ttsPlugin: {
             expectedSize: import("zod").ZodOptional<import("zod").ZodNumber>;
             sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
+                embeddings: "embeddings";
                 "llamacpp-completion": "llamacpp-completion";
                 "whispercpp-transcription": "whispercpp-transcription";
                 "llamacpp-embedding": "llamacpp-embedding";
@@ -183,7 +184,6 @@ export declare const ttsPlugin: {
                 "sdcpp-generation": "sdcpp-generation";
                 llm: "llm";
                 whisper: "whisper";
-                embeddings: "embeddings";
                 nmt: "nmt";
                 parakeet: "parakeet";
                 tts: "tts";
@@ -214,6 +214,7 @@ export declare const ttsPlugin: {
             expectedSize: import("zod").ZodOptional<import("zod").ZodNumber>;
             sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
+                embeddings: "embeddings";
                 "llamacpp-completion": "llamacpp-completion";
                 "whispercpp-transcription": "whispercpp-transcription";
                 "llamacpp-embedding": "llamacpp-embedding";
@@ -224,7 +225,6 @@ export declare const ttsPlugin: {
                 "sdcpp-generation": "sdcpp-generation";
                 llm: "llm";
                 whisper: "whisper";
-                embeddings: "embeddings";
                 nmt: "nmt";
                 parakeet: "parakeet";
                 tts: "tts";
@@ -244,6 +244,7 @@ export declare const ttsPlugin: {
             expectedSize: import("zod").ZodOptional<import("zod").ZodNumber>;
             sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
+                embeddings: "embeddings";
                 "llamacpp-completion": "llamacpp-completion";
                 "whispercpp-transcription": "whispercpp-transcription";
                 "llamacpp-embedding": "llamacpp-embedding";
@@ -254,7 +255,6 @@ export declare const ttsPlugin: {
                 "sdcpp-generation": "sdcpp-generation";
                 llm: "llm";
                 whisper: "whisper";
-                embeddings: "embeddings";
                 nmt: "nmt";
                 parakeet: "parakeet";
                 tts: "tts";
@@ -274,6 +274,7 @@ export declare const ttsPlugin: {
             expectedSize: import("zod").ZodOptional<import("zod").ZodNumber>;
             sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
+                embeddings: "embeddings";
                 "llamacpp-completion": "llamacpp-completion";
                 "whispercpp-transcription": "whispercpp-transcription";
                 "llamacpp-embedding": "llamacpp-embedding";
@@ -284,7 +285,6 @@ export declare const ttsPlugin: {
                 "sdcpp-generation": "sdcpp-generation";
                 llm: "llm";
                 whisper: "whisper";
-                embeddings: "embeddings";
                 nmt: "nmt";
                 parakeet: "parakeet";
                 tts: "tts";
@@ -304,6 +304,7 @@ export declare const ttsPlugin: {
             expectedSize: import("zod").ZodOptional<import("zod").ZodNumber>;
             sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
+                embeddings: "embeddings";
                 "llamacpp-completion": "llamacpp-completion";
                 "whispercpp-transcription": "whispercpp-transcription";
                 "llamacpp-embedding": "llamacpp-embedding";
@@ -314,7 +315,6 @@ export declare const ttsPlugin: {
                 "sdcpp-generation": "sdcpp-generation";
                 llm: "llm";
                 whisper: "whisper";
-                embeddings: "embeddings";
                 nmt: "nmt";
                 parakeet: "parakeet";
                 tts: "tts";
@@ -334,6 +334,7 @@ export declare const ttsPlugin: {
             expectedSize: import("zod").ZodOptional<import("zod").ZodNumber>;
             sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
+                embeddings: "embeddings";
                 "llamacpp-completion": "llamacpp-completion";
                 "whispercpp-transcription": "whispercpp-transcription";
                 "llamacpp-embedding": "llamacpp-embedding";
@@ -344,7 +345,6 @@ export declare const ttsPlugin: {
                 "sdcpp-generation": "sdcpp-generation";
                 llm: "llm";
                 whisper: "whisper";
-                embeddings: "embeddings";
                 nmt: "nmt";
                 parakeet: "parakeet";
                 tts: "tts";
@@ -364,6 +364,7 @@ export declare const ttsPlugin: {
             expectedSize: import("zod").ZodOptional<import("zod").ZodNumber>;
             sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
+                embeddings: "embeddings";
                 "llamacpp-completion": "llamacpp-completion";
                 "whispercpp-transcription": "whispercpp-transcription";
                 "llamacpp-embedding": "llamacpp-embedding";
@@ -374,7 +375,6 @@ export declare const ttsPlugin: {
                 "sdcpp-generation": "sdcpp-generation";
                 llm: "llm";
                 whisper: "whisper";
-                embeddings: "embeddings";
                 nmt: "nmt";
                 parakeet: "parakeet";
                 tts: "tts";
@@ -394,6 +394,7 @@ export declare const ttsPlugin: {
             expectedSize: import("zod").ZodOptional<import("zod").ZodNumber>;
             sha256Checksum: import("zod").ZodOptional<import("zod").ZodString>;
             addon: import("zod").ZodOptional<import("zod").ZodUnion<readonly [import("zod").ZodEnum<{
+                embeddings: "embeddings";
                 "llamacpp-completion": "llamacpp-completion";
                 "whispercpp-transcription": "whispercpp-transcription";
                 "llamacpp-embedding": "llamacpp-embedding";
@@ -404,7 +405,6 @@ export declare const ttsPlugin: {
                 "sdcpp-generation": "sdcpp-generation";
                 llm: "llm";
                 whisper: "whisper";
-                embeddings: "embeddings";
                 nmt: "nmt";
                 parakeet: "parakeet";
                 tts: "tts";
@@ -443,6 +443,9 @@ export declare const ttsPlugin: {
             inputType: import("zod").ZodDefault<import("zod").ZodString>;
             text: import("zod").ZodString;
             stream: import("zod").ZodDefault<import("zod").ZodBoolean>;
+            sentenceStream: import("zod").ZodDefault<import("zod").ZodBoolean>;
+            sentenceStreamLocale: import("zod").ZodOptional<import("zod").ZodString>;
+            sentenceStreamMaxChunkScalars: import("zod").ZodOptional<import("zod").ZodNumber>;
             type: import("zod").ZodLiteral<"textToSpeech">;
         }, import("zod/v4/core").$strip>, import("zod").ZodObject<{
             type: import("zod").ZodLiteral<"textToSpeech">;
@@ -452,6 +455,31 @@ export declare const ttsPlugin: {
                 audioDuration: import("zod").ZodOptional<import("zod").ZodNumber>;
                 totalSamples: import("zod").ZodOptional<import("zod").ZodNumber>;
             }, import("zod/v4/core").$strip>>;
+            chunkIndex: import("zod").ZodOptional<import("zod").ZodNumber>;
+            sentenceChunk: import("zod").ZodOptional<import("zod").ZodString>;
+        }, import("zod/v4/core").$strip>>;
+        textToSpeechStream: import("../../../../schemas/index").PluginHandlerDefinition<import("zod").ZodObject<{
+            modelId: import("zod").ZodString;
+            inputType: import("zod").ZodDefault<import("zod").ZodString>;
+            accumulateSentences: import("zod").ZodOptional<import("zod").ZodBoolean>;
+            sentenceDelimiterPreset: import("zod").ZodOptional<import("zod").ZodEnum<{
+                latin: "latin";
+                cjk: "cjk";
+                multilingual: "multilingual";
+            }>>;
+            maxBufferScalars: import("zod").ZodOptional<import("zod").ZodNumber>;
+            flushAfterMs: import("zod").ZodOptional<import("zod").ZodNumber>;
+            type: import("zod").ZodLiteral<"textToSpeechStream">;
+        }, import("zod/v4/core").$strip>, import("zod").ZodObject<{
+            type: import("zod").ZodLiteral<"textToSpeechStream">;
+            buffer: import("zod").ZodArray<import("zod").ZodNumber>;
+            done: import("zod").ZodDefault<import("zod").ZodBoolean>;
+            stats: import("zod").ZodOptional<import("zod").ZodObject<{
+                audioDuration: import("zod").ZodOptional<import("zod").ZodNumber>;
+                totalSamples: import("zod").ZodOptional<import("zod").ZodNumber>;
+            }, import("zod/v4/core").$strip>>;
+            chunkIndex: import("zod").ZodOptional<import("zod").ZodNumber>;
+            sentenceChunk: import("zod").ZodOptional<import("zod").ZodString>;
         }, import("zod/v4/core").$strip>>;
     };
     logging: {

package/dist/server/bare/plugins/onnx-tts/plugin.js

@@ -1,10 +1,11 @@
 import path from "bare-path";
 import ttsAddonLogging from "@qvac/tts-onnx/addonLogging.js";
 import ONNXTTS from "@qvac/tts-onnx";
-import { definePlugin, defineHandler, ttsRequestSchema, ttsResponseSchema, ModelType, ttsConfigSchema, ADDON_TTS, } from "../../../../schemas/index.js";
+import { definePlugin, defineHandler, defineDuplexHandler, ttsRequestSchema, ttsResponseSchema, textToSpeechStreamRequestSchema, textToSpeechStreamResponseSchema, ModelType, ttsConfigSchema, ADDON_TTS, } from "../../../../schemas/index.js";
 import { createStreamLogger, registerAddonLogger } from "../../../../logging/index.js";
 import { TtsArtifactsRequiredError, TtsReferenceAudioRequiredError, } from "../../../../utils/errors-server.js";
 import { textToSpeech } from "./ops/text-to-speech.js";
+import { textToSpeechStream } from "./ops/text-to-speech-stream.js";
 import { attachModelExecutionMs } from "../../../../profiling/model-execution.js";
 import { loadReferenceAudioAt24k } from "./wav-helper.js";
 async function resolveChatterboxConfig(config, ctx) {
@@ -118,7 +119,7 @@ function createChatterboxModel(modelId, config, artifacts) {
         opts: { stats: true },
         exclusiveRun: true,
     });
-    return { model, loader: undefined };
+    return { model };
 }
 function createSupertonicModel(modelId, config, artifacts) {
     const textEncoderPath = artifacts["textEncoderPath"];
@@ -160,7 +161,7 @@ function createSupertonicModel(modelId, config, artifacts) {
         opts: { stats: true },
         exclusiveRun: true,
     });
-    return { model, loader: undefined };
+    return { model };
 }
 export const ttsPlugin = definePlugin({
     modelType: ModelType.onnxTts,
@@ -197,6 +198,13 @@ export const ttsPlugin = definePlugin({
                             type: "textToSpeech",
                             buffer: result.value.buffer,
                             done: false,
+                            ...(result.value.chunkIndex !== undefined
+                                ? { chunkIndex: result.value.chunkIndex }
+                                : {}),
+                            ...(typeof result.value.sentenceChunk === "string" &&
+                                result.value.sentenceChunk.length > 0
+                                ? { sentenceChunk: result.value.sentenceChunk }
+                                : {}),
                         };
                         result = await stream.next();
                     }
@@ -213,6 +221,43 @@ export const ttsPlugin = definePlugin({
                 }
             },
         }),
+        textToSpeechStream: defineDuplexHandler({
+            requestSchema: textToSpeechStreamRequestSchema,
+            responseSchema: textToSpeechStreamResponseSchema,
+            streaming: true,
+            duplex: true,
+            handler: async function* (request, inputStream) {
+                const stream = textToSpeechStream(request, inputStream);
+                try {
+                    let result = await stream.next();
+                    while (!result.done) {
+                        yield {
+                            type: "textToSpeechStream",
+                            buffer: result.value.buffer,
+                            done: false,
+                            ...(result.value.chunkIndex !== undefined
+                                ? { chunkIndex: result.value.chunkIndex }
+                                : {}),
+                            ...(typeof result.value.sentenceChunk === "string" &&
+                                result.value.sentenceChunk.length > 0
+                                ? { sentenceChunk: result.value.sentenceChunk }
+                                : {}),
+                        };
+                        result = await stream.next();
+                    }
+                    const { modelExecutionMs, stats } = result.value;
+                    yield attachModelExecutionMs({
+                        type: "textToSpeechStream",
+                        buffer: [],
+                        done: true,
+                        ...(stats && { stats }),
+                    }, modelExecutionMs);
+                }
+                finally {
+                    await stream.return?.(undefined);
+                }
+            },
+        }),
     },
     logging: {
         module: ttsAddonLogging,