@livekit/agents-plugin-baseten 1.0.31

package/dist/stt.js

@@ -0,0 +1,246 @@
+import { AudioByteStream, Task, log, stt, waitForAbort } from "@livekit/agents";
+import { WebSocket } from "ws";
+const defaultSTTOptions = {
+  environment: "production",
+  encoding: "pcm_s16le",
+  sampleRate: 16e3,
+  bufferSizeSeconds: 0.032,
+  enablePartialTranscripts: true,
+  partialTranscriptIntervalS: 0.5,
+  finalTranscriptMaxDurationS: 5,
+  audioLanguage: "en",
+  languageDetectionOnly: false,
+  vadThreshold: 0.5,
+  vadMinSilenceDurationMs: 300,
+  vadSpeechPadMs: 30
+};
+class STT extends stt.STT {
+  #opts;
+  #logger = log();
+  label = "baseten.STT";
+  constructor(opts = {}) {
+    super({
+      streaming: true,
+      interimResults: opts.enablePartialTranscripts ?? defaultSTTOptions.enablePartialTranscripts
+    });
+    const apiKey = opts.apiKey ?? process.env.BASETEN_API_KEY;
+    const modelId = opts.modelId ?? process.env.BASETEN_STT_MODEL_ID;
+    if (!apiKey) {
+      throw new Error(
+        "Baseten API key is required, either pass it as `apiKey` or set $BASETEN_API_KEY"
+      );
+    }
+    if (!modelId) {
+      throw new Error(
+        "Baseten model ID is required, either pass it as `modelId` or set $BASETEN_STT_MODEL_ID"
+      );
+    }
+    this.#opts = {
+      ...defaultSTTOptions,
+      ...opts,
+      apiKey,
+      modelId
+    };
+  }
+  // eslint-disable-next-line
+  async _recognize(_) {
+    throw new Error("Recognize is not supported on Baseten STT");
+  }
+  updateOptions(opts) {
+    this.#opts = { ...this.#opts, ...opts };
+  }
+  stream() {
+    return new SpeechStream(this, this.#opts);
+  }
+}
+class SpeechStream extends stt.SpeechStream {
+  #opts;
+  #logger = log();
+  #speaking = false;
+  #requestId = "";
+  label = "baseten.SpeechStream";
+  constructor(stt2, opts) {
+    super(stt2, opts.sampleRate);
+    this.#opts = opts;
+    this.closed = false;
+  }
+  getWsUrl() {
+    return `wss://model-${this.#opts.modelId}.api.baseten.co/environments/${this.#opts.environment}/websocket`;
+  }
+  async run() {
+    const maxRetry = 32;
+    let retries = 0;
+    while (!this.input.closed && !this.closed) {
+      const url = this.getWsUrl();
+      const headers = {
+        Authorization: `Api-Key ${this.#opts.apiKey}`
+      };
+      const ws = new WebSocket(url, { headers });
+      try {
+        await new Promise((resolve, reject) => {
+          ws.on("open", resolve);
+          ws.on("error", (error) => reject(error));
+          ws.on("close", (code) => reject(`WebSocket returned ${code}`));
+        });
+        await this.#runWS(ws);
+      } catch (e) {
+        if (!this.closed && !this.input.closed) {
+          if (retries >= maxRetry) {
+            throw new Error(`failed to connect to Baseten after ${retries} attempts: ${e}`);
+          }
+          const delay = Math.min(retries * 5, 10);
+          retries++;
+          this.#logger.warn(
+            `failed to connect to Baseten, retrying in ${delay} seconds: ${e} (${retries}/${maxRetry})`
+          );
+          await new Promise((resolve) => setTimeout(resolve, delay * 1e3));
+        } else {
+          this.#logger.warn(
+            `Baseten disconnected, connection is closed: ${e} (inputClosed: ${this.input.closed}, isClosed: ${this.closed})`
+          );
+        }
+      }
+    }
+    this.closed = true;
+  }
+  async #runWS(ws) {
+    let closing = false;
+    const metadata = {
+      streaming_vad_config: {
+        threshold: this.#opts.vadThreshold,
+        min_silence_duration_ms: this.#opts.vadMinSilenceDurationMs,
+        speech_pad_ms: this.#opts.vadSpeechPadMs
+      },
+      streaming_params: {
+        encoding: this.#opts.encoding ?? "pcm_s16le",
+        sample_rate: this.#opts.sampleRate ?? 16e3,
+        enable_partial_transcripts: this.#opts.enablePartialTranscripts,
+        partial_transcript_interval_s: this.#opts.partialTranscriptIntervalS,
+        final_transcript_max_duration_s: this.#opts.finalTranscriptMaxDurationS
+      },
+      whisper_params: {
+        prompt: this.#opts.prompt,
+        audio_language: this.#opts.audioLanguage ?? "en",
+        language_detection_only: this.#opts.languageDetectionOnly ?? false
+      }
+    };
+    ws.send(JSON.stringify(metadata));
+    const sendTask = async () => {
+      const sampleRate = this.#opts.sampleRate ?? 16e3;
+      const samplesPerChunk = sampleRate === 16e3 ? 512 : 256;
+      const audioByteStream = new AudioByteStream(sampleRate, 1, samplesPerChunk);
+      try {
+        while (!this.closed) {
+          const result = await this.input.next();
+          if (result.done) {
+            break;
+          }
+          const data = result.value;
+          let frames;
+          if (data === SpeechStream.FLUSH_SENTINEL) {
+            frames = audioByteStream.flush();
+          } else {
+            if (data.sampleRate !== sampleRate || data.channels !== 1) {
+              throw new Error(
+                `sample rate or channel count mismatch: expected ${sampleRate}Hz/1ch, got ${data.sampleRate}Hz/${data.channels}ch`
+              );
+            }
+            frames = audioByteStream.write(data.data.buffer);
+          }
+          for (const frame of frames) {
+            const buffer = Buffer.from(
+              frame.data.buffer,
+              frame.data.byteOffset,
+              frame.data.byteLength
+            );
+            ws.send(buffer);
+          }
+        }
+      } finally {
+        closing = true;
+        ws.close();
+      }
+    };
+    const listenTask = Task.from(async (controller) => {
+      const listenMessage = new Promise((resolve, reject) => {
+        ws.on("message", (data) => {
+          try {
+            let jsonString;
+            if (typeof data === "string") {
+              jsonString = data;
+            } else if (data instanceof Buffer) {
+              jsonString = data.toString("utf-8");
+            } else if (Array.isArray(data)) {
+              jsonString = Buffer.concat(data).toString("utf-8");
+            } else {
+              return;
+            }
+            const msg = JSON.parse(jsonString);
+            const isFinal = msg.is_final ?? true;
+            const segments = msg.segments ?? [];
+            const transcript = msg.transcript ?? "";
+            const confidence = msg.confidence ?? 0;
+            const languageCode = msg.language_code ?? this.#opts.audioLanguage;
+            if (!transcript) {
+              this.#logger.debug("Received non-transcript message:", msg);
+              return;
+            }
+            if (!this.#speaking && !isFinal) {
+              this.#speaking = true;
+              this.queue.put({ type: stt.SpeechEventType.START_OF_SPEECH });
+            }
+            const startTime = segments.length > 0 ? segments[0].start ?? 0 : 0;
+            const endTime = segments.length > 0 ? segments[segments.length - 1].end ?? 0 : 0;
+            const speechData = {
+              language: languageCode,
+              text: transcript,
+              startTime,
+              endTime,
+              confidence
+            };
+            if (!isFinal) {
+              this.queue.put({
+                type: stt.SpeechEventType.INTERIM_TRANSCRIPT,
+                alternatives: [speechData]
+              });
+            } else {
+              this.queue.put({
+                type: stt.SpeechEventType.FINAL_TRANSCRIPT,
+                alternatives: [speechData]
+              });
+              if (this.#speaking) {
+                this.#speaking = false;
+                this.queue.put({ type: stt.SpeechEventType.END_OF_SPEECH });
+              }
+            }
+            if (this.closed || closing) {
+              resolve();
+            }
+          } catch (err) {
+            this.#logger.error(`STT: Error processing message: ${data}`);
+            reject(err);
+          }
+        });
+        ws.on("error", (err) => {
+          if (!closing) {
+            reject(err);
+          }
+        });
+        ws.on("close", () => {
+          if (!closing) {
+            resolve();
+          }
+        });
+      });
+      await Promise.race([listenMessage, waitForAbort(controller.signal)]);
+    }, this.abortController);
+    await Promise.all([sendTask(), listenTask.result]);
+    closing = true;
+    ws.close();
+  }
+}
+export {
+  STT,
+  SpeechStream
+};
+//# sourceMappingURL=stt.js.map

package/dist/stt.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/stt.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2024 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport { type AudioBuffer, AudioByteStream, Task, log, stt, waitForAbort } from '@livekit/agents';\nimport type { AudioFrame } from '@livekit/rtc-node';\nimport { WebSocket } from 'ws';\nimport type { BasetenSttOptions } from './types.js';\n\nconst defaultSTTOptions: Partial<BasetenSttOptions> = {\n  environment: 'production',\n  encoding: 'pcm_s16le',\n  sampleRate: 16000,\n  bufferSizeSeconds: 0.032,\n  enablePartialTranscripts: true,\n  partialTranscriptIntervalS: 0.5,\n  finalTranscriptMaxDurationS: 5,\n  audioLanguage: 'en',\n  languageDetectionOnly: false,\n  vadThreshold: 0.5,\n  vadMinSilenceDurationMs: 300,\n  vadSpeechPadMs: 30,\n};\n\nexport class STT extends stt.STT {\n  #opts: BasetenSttOptions;\n  #logger = log();\n  label = 'baseten.STT';\n\n  constructor(opts: Partial<BasetenSttOptions> = {}) {\n    super({\n      streaming: true,\n      interimResults: opts.enablePartialTranscripts ?? defaultSTTOptions.enablePartialTranscripts!,\n    });\n\n    const apiKey = opts.apiKey ?? process.env.BASETEN_API_KEY;\n    const modelId = opts.modelId ?? process.env.BASETEN_STT_MODEL_ID;\n\n    if (!apiKey) {\n      throw new Error(\n        'Baseten API key is required, either pass it as `apiKey` or set $BASETEN_API_KEY',\n      );\n    }\n    if (!modelId) {\n      throw new Error(\n        'Baseten model ID is required, either pass it as `modelId` or set $BASETEN_STT_MODEL_ID',\n      );\n    }\n\n    this.#opts = {\n      ...defaultSTTOptions,\n      ...opts,\n      apiKey,\n      modelId,\n    } as BasetenSttOptions;\n  }\n\n  // eslint-disable-next-line\n  async _recognize(_: AudioBuffer): Promise<stt.SpeechEvent> {\n    throw new Error('Recognize is not supported on Baseten STT');\n  }\n\n  updateOptions(opts: Partial<BasetenSttOptions>) {\n    this.#opts = { ...this.#opts, ...opts };\n  }\n\n  stream(): SpeechStream {\n    return new SpeechStream(this, this.#opts);\n  }\n}\n\nexport class SpeechStream extends stt.SpeechStream {\n  #opts: BasetenSttOptions;\n  #logger = log();\n  #speaking = false;\n  #requestId = '';\n  label = 'baseten.SpeechStream';\n\n  constructor(stt: STT, opts: BasetenSttOptions) {\n    super(stt, opts.sampleRate);\n    this.#opts = opts;\n    this.closed = false;\n  }\n\n  private getWsUrl(): string {\n    return `wss://model-${this.#opts.modelId}.api.baseten.co/environments/${this.#opts.environment}/websocket`;\n  }\n\n  protected async run() {\n    const maxRetry = 32;\n    let retries = 0;\n\n    while (!this.input.closed && !this.closed) {\n      const url = this.getWsUrl();\n      const headers = {\n        Authorization: `Api-Key ${this.#opts.apiKey}`,\n      };\n\n      const ws = new WebSocket(url, { headers });\n\n      try {\n        await new Promise((resolve, reject) => {\n          ws.on('open', resolve);\n          ws.on('error', (error) => reject(error));\n          ws.on('close', (code) => reject(`WebSocket returned ${code}`));\n        });\n\n        await this.#runWS(ws);\n      } catch (e) {\n        if (!this.closed && !this.input.closed) {\n          if (retries >= maxRetry) {\n            throw new Error(`failed to connect to Baseten after ${retries} attempts: ${e}`);\n          }\n\n          const delay = Math.min(retries * 5, 10);\n          retries++;\n\n          this.#logger.warn(\n            `failed to connect to Baseten, retrying in ${delay} seconds: ${e} (${retries}/${maxRetry})`,\n          );\n          await new Promise((resolve) => setTimeout(resolve, delay * 1000));\n        } else {\n          this.#logger.warn(\n            `Baseten disconnected, connection is closed: ${e} (inputClosed: ${this.input.closed}, isClosed: ${this.closed})`,\n          );\n        }\n      }\n    }\n\n    this.closed = true;\n  }\n\n  async #runWS(ws: WebSocket) {\n    let closing = false;\n\n    // Send initial metadata\n    const metadata = {\n      streaming_vad_config: {\n        threshold: this.#opts.vadThreshold,\n        min_silence_duration_ms: this.#opts.vadMinSilenceDurationMs,\n        speech_pad_ms: this.#opts.vadSpeechPadMs,\n      },\n      streaming_params: {\n        encoding: this.#opts.encoding ?? 'pcm_s16le',\n        sample_rate: this.#opts.sampleRate ?? 16000,\n        enable_partial_transcripts: this.#opts.enablePartialTranscripts,\n        partial_transcript_interval_s: this.#opts.partialTranscriptIntervalS,\n        final_transcript_max_duration_s: this.#opts.finalTranscriptMaxDurationS,\n      },\n      whisper_params: {\n        prompt: this.#opts.prompt,\n        audio_language: this.#opts.audioLanguage ?? 'en',\n        language_detection_only: this.#opts.languageDetectionOnly ?? false,\n      },\n    };\n    ws.send(JSON.stringify(metadata));\n\n    const sendTask = async () => {\n      const sampleRate = this.#opts.sampleRate ?? 16000;\n      const samplesPerChunk = sampleRate === 16000 ? 512 : 256;\n      const audioByteStream = new AudioByteStream(sampleRate, 1, samplesPerChunk);\n\n      try {\n        while (!this.closed) {\n          const result = await this.input.next();\n          if (result.done) {\n            break;\n          }\n\n          const data = result.value;\n\n          let frames: AudioFrame[];\n          if (data === SpeechStream.FLUSH_SENTINEL) {\n            // Flush any remaining buffered audio\n            frames = audioByteStream.flush();\n          } else {\n            if (data.sampleRate !== sampleRate || data.channels !== 1) {\n              throw new Error(\n                `sample rate or channel count mismatch: expected ${sampleRate}Hz/1ch, got ${data.sampleRate}Hz/${data.channels}ch`,\n              );\n            }\n            frames = audioByteStream.write(data.data.buffer as ArrayBuffer);\n          }\n\n          for (const frame of frames) {\n            const buffer = Buffer.from(\n              frame.data.buffer,\n              frame.data.byteOffset,\n              frame.data.byteLength,\n            );\n            ws.send(buffer);\n          }\n        }\n      } finally {\n        closing = true;\n        ws.close();\n      }\n    };\n\n    const listenTask = Task.from(async (controller) => {\n      const listenMessage = new Promise<void>((resolve, reject) => {\n        ws.on('message', (data) => {\n          try {\n            let jsonString: string;\n\n            if (typeof data === 'string') {\n              jsonString = data;\n            } else if (data instanceof Buffer) {\n              jsonString = data.toString('utf-8');\n            } else if (Array.isArray(data)) {\n              jsonString = Buffer.concat(data).toString('utf-8');\n            } else {\n              return;\n            }\n\n            const msg = JSON.parse(jsonString);\n\n            // Parse response format matching Python implementation\n            const isFinal = msg.is_final ?? true;\n            const segments = msg.segments ?? [];\n            const transcript = msg.transcript ?? '';\n            const confidence = msg.confidence ?? 0.0;\n            const languageCode = msg.language_code ?? this.#opts.audioLanguage;\n\n            // Skip if no transcript text\n            if (!transcript) {\n              this.#logger.debug('Received non-transcript message:', msg);\n              return;\n            }\n\n            // Emit START_OF_SPEECH if not already speaking (only for interim or first final)\n            if (!this.#speaking && !isFinal) {\n              this.#speaking = true;\n              this.queue.put({ type: stt.SpeechEventType.START_OF_SPEECH });\n            }\n\n            // Extract timing from segments\n            const startTime = segments.length > 0 ? segments[0].start ?? 0.0 : 0.0;\n            const endTime = segments.length > 0 ? segments[segments.length - 1].end ?? 0.0 : 0.0;\n\n            const speechData: stt.SpeechData = {\n              language: languageCode!,\n              text: transcript,\n              startTime,\n              endTime,\n              confidence,\n            };\n\n            // Handle interim vs final transcripts (matching Python implementation)\n            if (!isFinal) {\n              // Interim transcript\n              this.queue.put({\n                type: stt.SpeechEventType.INTERIM_TRANSCRIPT,\n                alternatives: [speechData],\n              });\n            } else {\n              // Final transcript\n              this.queue.put({\n                type: stt.SpeechEventType.FINAL_TRANSCRIPT,\n                alternatives: [speechData],\n              });\n\n              // Emit END_OF_SPEECH after final transcript\n              if (this.#speaking) {\n                this.#speaking = false;\n                this.queue.put({ type: stt.SpeechEventType.END_OF_SPEECH });\n              }\n            }\n\n            if (this.closed || closing) {\n              resolve();\n            }\n          } catch (err) {\n            this.#logger.error(`STT: Error processing message: ${data}`);\n            reject(err);\n          }\n        });\n\n        ws.on('error', (err) => {\n          if (!closing) {\n            reject(err);\n          }\n        });\n\n        ws.on('close', () => {\n          if (!closing) {\n            resolve();\n          }\n        });\n      });\n\n      await Promise.race([listenMessage, waitForAbort(controller.signal)]);\n    }, this.abortController);\n\n    await Promise.all([sendTask(), listenTask.result]);\n    closing = true;\n    ws.close();\n  }\n}\n"],"mappings":"AAGA,SAA2B,iBAAiB,MAAM,KAAK,KAAK,oBAAoB;AAEhF,SAAS,iBAAiB;AAG1B,MAAM,oBAAgD;AAAA,EACpD,aAAa;AAAA,EACb,UAAU;AAAA,EACV,YAAY;AAAA,EACZ,mBAAmB;AAAA,EACnB,0BAA0B;AAAA,EAC1B,4BAA4B;AAAA,EAC5B,6BAA6B;AAAA,EAC7B,eAAe;AAAA,EACf,uBAAuB;AAAA,EACvB,cAAc;AAAA,EACd,yBAAyB;AAAA,EACzB,gBAAgB;AAClB;AAEO,MAAM,YAAY,IAAI,IAAI;AAAA,EAC/B;AAAA,EACA,UAAU,IAAI;AAAA,EACd,QAAQ;AAAA,EAER,YAAY,OAAmC,CAAC,GAAG;AACjD,UAAM;AAAA,MACJ,WAAW;AAAA,MACX,gBAAgB,KAAK,4BAA4B,kBAAkB;AAAA,IACrE,CAAC;AAED,UAAM,SAAS,KAAK,UAAU,QAAQ,IAAI;AAC1C,UAAM,UAAU,KAAK,WAAW,QAAQ,IAAI;AAE5C,QAAI,CAAC,QAAQ;AACX,YAAM,IAAI;AAAA,QACR;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,SAAS;AACZ,YAAM,IAAI;AAAA,QACR;AAAA,MACF;AAAA,IACF;AAEA,SAAK,QAAQ;AAAA,MACX,GAAG;AAAA,MACH,GAAG;AAAA,MACH;AAAA,MACA;AAAA,IACF;AAAA,EACF;AAAA;AAAA,EAGA,MAAM,WAAW,GAA0C;AACzD,UAAM,IAAI,MAAM,2CAA2C;AAAA,EAC7D;AAAA,EAEA,cAAc,MAAkC;AAC9C,SAAK,QAAQ,EAAE,GAAG,KAAK,OAAO,GAAG,KAAK;AAAA,EACxC;AAAA,EAEA,SAAuB;AACrB,WAAO,IAAI,aAAa,MAAM,KAAK,KAAK;AAAA,EAC1C;AACF;AAEO,MAAM,qBAAqB,IAAI,aAAa;AAAA,EACjD;AAAA,EACA,UAAU,IAAI;AAAA,EACd,YAAY;AAAA,EACZ,aAAa;AAAA,EACb,QAAQ;AAAA,EAER,YAAYA,MAAU,MAAyB;AAC7C,UAAMA,MAAK,KAAK,UAAU;AAC1B,SAAK,QAAQ;AACb,SAAK,SAAS;AAAA,EAChB;AAAA,EAEQ,WAAmB;AACzB,WAAO,eAAe,KAAK,MAAM,OAAO,gCAAgC,KAAK,MAAM,WAAW;AAAA,EAChG;AAAA,EAEA,MAAgB,MAAM;AACpB,UAAM,WAAW;AACjB,QAAI,UAAU;AAEd,WAAO,CAAC,KAAK,MAAM,UAAU,CAAC,KAAK,QAAQ;AACzC,YAAM,MAAM,KAAK,SAAS;AAC1B,YAAM,UAAU;AAAA,QACd,eAAe,WAAW,KAAK,MAAM,MAAM;AAAA,MAC7C;AAEA,YAAM,KAAK,IAAI,UAAU,KAAK,EAAE,QAAQ,CAAC;AAEzC,UAAI;AACF,cAAM,IAAI,QAAQ,CAAC,SAAS,WAAW;AACrC,aAAG,GAAG,QAAQ,OAAO;AACrB,aAAG,GAAG,SAAS,CAAC,UAAU,OAAO,KAAK,CAAC;AACvC,aAAG,GAAG,SAAS,CAAC,SAAS,OAAO,sBAAsB,IAAI,EAAE,CAAC;AAAA,QAC/D,CAAC;AAED,cAAM,KAAK,OAAO,EAAE;AAAA,MACtB,SAAS,GAAG;AACV,YAAI,CAAC,KAAK,UAAU,CAAC,KAAK,MAAM,QAAQ;AACtC,cAAI,WAAW,UAAU;AACvB,kBAAM,IAAI,MAAM,sCAAsC,OAAO,cAAc,CAAC,EAAE;AAAA,UAChF;AAEA,gBAAM,QAAQ,KAAK,IAAI,UAAU,GAAG,EAAE;AACtC;AAEA,eAAK,QAAQ;AAAA,YACX,6CAA6C,KAAK,aAAa,CAAC,KAAK,OAAO,IAAI,QAAQ;AAAA,UAC1F;AACA,gBAAM,IAAI,QAAQ,CAAC,YAAY,WAAW,SAAS,QAAQ,GAAI,CAAC;AAAA,QAClE,OAAO;AACL,eAAK,QAAQ;AAAA,YACX,+CAA+C,CAAC,kBAAkB,KAAK,MAAM,MAAM,eAAe,KAAK,MAAM;AAAA,UAC/G;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAEA,SAAK,SAAS;AAAA,EAChB;AAAA,EAEA,MAAM,OAAO,IAAe;AAC1B,QAAI,UAAU;AAGd,UAAM,WAAW;AAAA,MACf,sBAAsB;AAAA,QACpB,WAAW,KAAK,MAAM;AAAA,QACtB,yBAAyB,KAAK,MAAM;AAAA,QACpC,eAAe,KAAK,MAAM;AAAA,MAC5B;AAAA,MACA,kBAAkB;AAAA,QAChB,UAAU,KAAK,MAAM,YAAY;AAAA,QACjC,aAAa,KAAK,MAAM,cAAc;AAAA,QACtC,4BAA4B,KAAK,MAAM;AAAA,QACvC,+BAA+B,KAAK,MAAM;AAAA,QAC1C,iCAAiC,KAAK,MAAM;AAAA,MAC9C;AAAA,MACA,gBAAgB;AAAA,QACd,QAAQ,KAAK,MAAM;AAAA,QACnB,gBAAgB,KAAK,MAAM,iBAAiB;AAAA,QAC5C,yBAAyB,KAAK,MAAM,yBAAyB;AAAA,MAC/D;AAAA,IACF;AACA,OAAG,KAAK,KAAK,UAAU,QAAQ,CAAC;AAEhC,UAAM,WAAW,YAAY;AAC3B,YAAM,aAAa,KAAK,MAAM,cAAc;AAC5C,YAAM,kBAAkB,eAAe,OAAQ,MAAM;AACrD,YAAM,kBAAkB,IAAI,gBAAgB,YAAY,GAAG,eAAe;AAE1E,UAAI;AACF,eAAO,CAAC,KAAK,QAAQ;AACnB,gBAAM,SAAS,MAAM,KAAK,MAAM,KAAK;AACrC,cAAI,OAAO,MAAM;AACf;AAAA,UACF;AAEA,gBAAM,OAAO,OAAO;AAEpB,cAAI;AACJ,cAAI,SAAS,aAAa,gBAAgB;AAExC,qBAAS,gBAAgB,MAAM;AAAA,UACjC,OAAO;AACL,gBAAI,KAAK,eAAe,cAAc,KAAK,aAAa,GAAG;AACzD,oBAAM,IAAI;AAAA,gBACR,mDAAmD,UAAU,eAAe,KAAK,UAAU,MAAM,KAAK,QAAQ;AAAA,cAChH;AAAA,YACF;AACA,qBAAS,gBAAgB,MAAM,KAAK,KAAK,MAAqB;AAAA,UAChE;AAEA,qBAAW,SAAS,QAAQ;AAC1B,kBAAM,SAAS,OAAO;AAAA,cACpB,MAAM,KAAK;AAAA,cACX,MAAM,KAAK;AAAA,cACX,MAAM,KAAK;AAAA,YACb;AACA,eAAG,KAAK,MAAM;AAAA,UAChB;AAAA,QACF;AAAA,MACF,UAAE;AACA,kBAAU;AACV,WAAG,MAAM;AAAA,MACX;AAAA,IACF;AAEA,UAAM,aAAa,KAAK,KAAK,OAAO,eAAe;AACjD,YAAM,gBAAgB,IAAI,QAAc,CAAC,SAAS,WAAW;AAC3D,WAAG,GAAG,WAAW,CAAC,SAAS;AACzB,cAAI;AACF,gBAAI;AAEJ,gBAAI,OAAO,SAAS,UAAU;AAC5B,2BAAa;AAAA,YACf,WAAW,gBAAgB,QAAQ;AACjC,2BAAa,KAAK,SAAS,OAAO;AAAA,YACpC,WAAW,MAAM,QAAQ,IAAI,GAAG;AAC9B,2BAAa,OAAO,OAAO,IAAI,EAAE,SAAS,OAAO;AAAA,YACnD,OAAO;AACL;AAAA,YACF;AAEA,kBAAM,MAAM,KAAK,MAAM,UAAU;AAGjC,kBAAM,UAAU,IAAI,YAAY;AAChC,kBAAM,WAAW,IAAI,YAAY,CAAC;AAClC,kBAAM,aAAa,IAAI,cAAc;AACrC,kBAAM,aAAa,IAAI,cAAc;AACrC,kBAAM,eAAe,IAAI,iBAAiB,KAAK,MAAM;AAGrD,gBAAI,CAAC,YAAY;AACf,mBAAK,QAAQ,MAAM,oCAAoC,GAAG;AAC1D;AAAA,YACF;AAGA,gBAAI,CAAC,KAAK,aAAa,CAAC,SAAS;AAC/B,mBAAK,YAAY;AACjB,mBAAK,MAAM,IAAI,EAAE,MAAM,IAAI,gBAAgB,gBAAgB,CAAC;AAAA,YAC9D;AAGA,kBAAM,YAAY,SAAS,SAAS,IAAI,SAAS,CAAC,EAAE,SAAS,IAAM;AACnE,kBAAM,UAAU,SAAS,SAAS,IAAI,SAAS,SAAS,SAAS,CAAC,EAAE,OAAO,IAAM;AAEjF,kBAAM,aAA6B;AAAA,cACjC,UAAU;AAAA,cACV,MAAM;AAAA,cACN;AAAA,cACA;AAAA,cACA;AAAA,YACF;AAGA,gBAAI,CAAC,SAAS;AAEZ,mBAAK,MAAM,IAAI;AAAA,gBACb,MAAM,IAAI,gBAAgB;AAAA,gBAC1B,cAAc,CAAC,UAAU;AAAA,cAC3B,CAAC;AAAA,YACH,OAAO;AAEL,mBAAK,MAAM,IAAI;AAAA,gBACb,MAAM,IAAI,gBAAgB;AAAA,gBAC1B,cAAc,CAAC,UAAU;AAAA,cAC3B,CAAC;AAGD,kBAAI,KAAK,WAAW;AAClB,qBAAK,YAAY;AACjB,qBAAK,MAAM,IAAI,EAAE,MAAM,IAAI,gBAAgB,cAAc,CAAC;AAAA,cAC5D;AAAA,YACF;AAEA,gBAAI,KAAK,UAAU,SAAS;AAC1B,sBAAQ;AAAA,YACV;AAAA,UACF,SAAS,KAAK;AACZ,iBAAK,QAAQ,MAAM,kCAAkC,IAAI,EAAE;AAC3D,mBAAO,GAAG;AAAA,UACZ;AAAA,QACF,CAAC;AAED,WAAG,GAAG,SAAS,CAAC,QAAQ;AACtB,cAAI,CAAC,SAAS;AACZ,mBAAO,GAAG;AAAA,UACZ;AAAA,QACF,CAAC;AAED,WAAG,GAAG,SAAS,MAAM;AACnB,cAAI,CAAC,SAAS;AACZ,oBAAQ;AAAA,UACV;AAAA,QACF,CAAC;AAAA,MACH,CAAC;AAED,YAAM,QAAQ,KAAK,CAAC,eAAe,aAAa,WAAW,MAAM,CAAC,CAAC;AAAA,IACrE,GAAG,KAAK,eAAe;AAEvB,UAAM,QAAQ,IAAI,CAAC,SAAS,GAAG,WAAW,MAAM,CAAC;AACjD,cAAU;AACV,OAAG,MAAM;AAAA,EACX;AACF;","names":["stt"]}

package/dist/stt.test.cjs

@@ -0,0 +1,9 @@
+"use strict";
+var import_agents_plugin_silero = require("@livekit/agents-plugin-silero");
+var import_agents_plugins_test = require("@livekit/agents-plugins-test");
+var import_vitest = require("vitest");
+var import_stt = require("./stt.cjs");
+(0, import_vitest.describe)("Baseten", async () => {
+  await (0, import_agents_plugins_test.stt)(new import_stt.STT(), await import_agents_plugin_silero.VAD.load(), { streaming: true });
+});
+//# sourceMappingURL=stt.test.cjs.map

package/dist/stt.test.cjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/stt.test.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2024 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport { VAD } from '@livekit/agents-plugin-silero';\nimport { stt } from '@livekit/agents-plugins-test';\nimport { describe } from 'vitest';\nimport { STT } from './stt.js';\n\ndescribe('Baseten', async () => {\n  await stt(new STT(), await VAD.load(), { streaming: true });\n});\n"],"mappings":";AAGA,kCAAoB;AACpB,iCAAoB;AACpB,oBAAyB;AACzB,iBAAoB;AAAA,IAEpB,wBAAS,WAAW,YAAY;AAC9B,YAAM,gCAAI,IAAI,eAAI,GAAG,MAAM,gCAAI,KAAK,GAAG,EAAE,WAAW,KAAK,CAAC;AAC5D,CAAC;","names":[]}

package/dist/stt.test.d.cts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=stt.test.d.ts.map

package/dist/stt.test.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=stt.test.d.ts.map

package/dist/stt.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"stt.test.d.ts","sourceRoot":"","sources":["../src/stt.test.ts"],"names":[],"mappings":""}

package/dist/stt.test.js

@@ -0,0 +1,8 @@
+import { VAD } from "@livekit/agents-plugin-silero";
+import { stt } from "@livekit/agents-plugins-test";
+import { describe } from "vitest";
+import { STT } from "./stt.js";
+describe("Baseten", async () => {
+  await stt(new STT(), await VAD.load(), { streaming: true });
+});
+//# sourceMappingURL=stt.test.js.map

package/dist/stt.test.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/stt.test.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2024 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport { VAD } from '@livekit/agents-plugin-silero';\nimport { stt } from '@livekit/agents-plugins-test';\nimport { describe } from 'vitest';\nimport { STT } from './stt.js';\n\ndescribe('Baseten', async () => {\n  await stt(new STT(), await VAD.load(), { streaming: true });\n});\n"],"mappings":"AAGA,SAAS,WAAW;AACpB,SAAS,WAAW;AACpB,SAAS,gBAAgB;AACzB,SAAS,WAAW;AAEpB,SAAS,WAAW,YAAY;AAC9B,QAAM,IAAI,IAAI,IAAI,GAAG,MAAM,IAAI,KAAK,GAAG,EAAE,WAAW,KAAK,CAAC;AAC5D,CAAC;","names":[]}

package/dist/tts.cjs

@@ -0,0 +1,161 @@
+"use strict";
+var __defProp = Object.defineProperty;
+var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
+var __getOwnPropNames = Object.getOwnPropertyNames;
+var __hasOwnProp = Object.prototype.hasOwnProperty;
+var __export = (target, all) => {
+  for (var name in all)
+    __defProp(target, name, { get: all[name], enumerable: true });
+};
+var __copyProps = (to, from, except, desc) => {
+  if (from && typeof from === "object" || typeof from === "function") {
+    for (let key of __getOwnPropNames(from))
+      if (!__hasOwnProp.call(to, key) && key !== except)
+        __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
+  }
+  return to;
+};
+var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
+var tts_exports = {};
+__export(tts_exports, {
+  ChunkedStream: () => ChunkedStream,
+  TTS: () => TTS
+});
+module.exports = __toCommonJS(tts_exports);
+var import_agents = require("@livekit/agents");
+const defaultTTSOptions = {
+  voice: "tara",
+  language: "en",
+  temperature: 0.6
+};
+class TTS extends import_agents.tts.TTS {
+  opts;
+  label = "baseten.TTS";
+  abortController = new AbortController();
+  constructor(opts = {}) {
+    super(24e3, 1, { streaming: false });
+    const apiKey = opts.apiKey ?? process.env.BASETEN_API_KEY;
+    const modelEndpoint = opts.modelEndpoint ?? process.env.BASETEN_MODEL_ENDPOINT;
+    if (!apiKey) {
+      throw new Error(
+        "Baseten API key is required, either pass it as `apiKey` or set $BASETEN_API_KEY"
+      );
+    }
+    if (!modelEndpoint) {
+      throw new Error(
+        "Baseten model endpoint is required, either pass it as `modelEndpoint` or set $BASETEN_MODEL_ENDPOINT"
+      );
+    }
+    this.opts = {
+      ...defaultTTSOptions,
+      ...opts,
+      apiKey,
+      modelEndpoint
+    };
+  }
+  updateOptions(opts) {
+    this.opts = {
+      ...this.opts,
+      ...opts
+    };
+  }
+  /**
+   * Synthesize speech for a given piece of text.  Returns a `ChunkedStream`
+   * which will asynchronously fetch audio from Baseten and push frames into
+   * LiveKit's playback pipeline.  If you need to cancel synthesis you can
+   * call {@link ChunkedStream.stop} on the returned object.
+   */
+  synthesize(text, connOptions, abortSignal) {
+    return new ChunkedStream(this, text, this.opts, connOptions, abortSignal);
+  }
+  stream() {
+    throw new Error("Streaming is not supported on Baseten TTS");
+  }
+  async close() {
+    this.abortController.abort();
+  }
+}
+class ChunkedStream extends import_agents.tts.ChunkedStream {
+  label = "baseten.ChunkedStream";
+  opts;
+  constructor(tts2, text, opts, connOptions, abortSignal) {
+    super(text, tts2, connOptions, abortSignal);
+    this.opts = opts;
+  }
+  /**
+   * Execute the synthesis request.  This method is automatically invoked
+   * by the base class when the stream starts.  It performs a POST request
+   * to the configured `modelEndpoint` with the input text and optional
+   * parameters.  Audio chunks are streamed as they arrive and transformed
+   * into a sequence of `AudioFrame` objects that are enqueued immediately
+   * for playback.
+   */
+  async run() {
+    const { apiKey, modelEndpoint, voice, language, temperature, maxTokens } = this.opts;
+    const payload = {
+      prompt: this.inputText
+    };
+    if (voice) payload.voice = voice;
+    if (language) payload.language = language;
+    if (temperature !== void 0) payload.temperature = temperature;
+    if (maxTokens !== void 0) payload.max_tokens = maxTokens;
+    const headers = {
+      Authorization: `Api-Key ${apiKey}`,
+      "Content-Type": "application/json"
+    };
+    const response = await fetch(modelEndpoint, {
+      method: "POST",
+      headers,
+      body: JSON.stringify(payload),
+      signal: this.abortSignal
+    });
+    if (!response.ok) {
+      let errText;
+      try {
+        errText = await response.text();
+      } catch {
+        errText = response.statusText;
+      }
+      throw new Error(`Baseten TTS request failed: ${response.status} ${errText}`);
+    }
+    if (!response.body) {
+      throw new Error("Response body is not available for streaming");
+    }
+    const requestId = (0, import_agents.shortuuid)();
+    const audioByteStream = new import_agents.AudioByteStream(24e3, 1);
+    const reader = response.body.getReader();
+    try {
+      let lastFrame;
+      const sendLastFrame = (segmentId, final) => {
+        if (lastFrame) {
+          this.queue.put({ requestId, segmentId, frame: lastFrame, final });
+          lastFrame = void 0;
+        }
+      };
+      const abortPromise = (0, import_agents.waitForAbort)(this.abortSignal);
+      while (!this.abortSignal.aborted) {
+        const result = await Promise.race([reader.read(), abortPromise]);
+        if (result === void 0) break;
+        const { done, value } = result;
+        if (done) {
+          break;
+        }
+        const frames = audioByteStream.write(value.buffer);
+        for (const frame of frames) {
+          sendLastFrame(requestId, false);
+          lastFrame = frame;
+        }
+      }
+      sendLastFrame(requestId, true);
+    } finally {
+      reader.releaseLock();
+      this.queue.close();
+    }
+  }
+}
+// Annotate the CommonJS export names for ESM import in node:
+0 && (module.exports = {
+  ChunkedStream,
+  TTS
+});
+//# sourceMappingURL=tts.cjs.map

package/dist/tts.cjs.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/tts.ts"],"sourcesContent":["// SPDX-FileCopyrightText: 2024 LiveKit, Inc.\n//\n// SPDX-License-Identifier: Apache-2.0\nimport {\n  type APIConnectOptions,\n  AudioByteStream,\n  shortuuid,\n  tts,\n  waitForAbort,\n} from '@livekit/agents';\nimport type { AudioFrame } from '@livekit/rtc-node';\nimport type { BasetenTTSOptions } from './types.js';\n\nconst defaultTTSOptions: Partial<BasetenTTSOptions> = {\n  voice: 'tara',\n  language: 'en',\n  temperature: 0.6,\n};\n\n/**\n * Baseten TTS implementation (streaming, 24kHz mono)\n */\nexport class TTS extends tts.TTS {\n  private opts: BasetenTTSOptions;\n  label = 'baseten.TTS';\n  private abortController = new AbortController();\n  constructor(opts: Partial<BasetenTTSOptions> = {}) {\n    /**\n     * Baseten audio is 24kHz mono.\n     * The Orpheus model generates audio chunks that are processed as they arrive,\n     * which reduces latency and improves agent responsiveness.\n     */\n    super(24000, 1, { streaming: false });\n\n    // Apply defaults and environment fallbacks.\n    const apiKey = opts.apiKey ?? process.env.BASETEN_API_KEY;\n    const modelEndpoint = opts.modelEndpoint ?? process.env.BASETEN_MODEL_ENDPOINT;\n\n    if (!apiKey) {\n      throw new Error(\n        'Baseten API key is required, either pass it as `apiKey` or set $BASETEN_API_KEY',\n      );\n    }\n    if (!modelEndpoint) {\n      throw new Error(\n        'Baseten model endpoint is required, either pass it as `modelEndpoint` or set $BASETEN_MODEL_ENDPOINT',\n      );\n    }\n\n    this.opts = {\n      ...defaultTTSOptions,\n      ...opts,\n      apiKey,\n      modelEndpoint,\n    } as BasetenTTSOptions;\n  }\n\n  updateOptions(opts: Partial<Omit<BasetenTTSOptions, 'apiKey' | 'modelEndpoint'>>) {\n    this.opts = {\n      ...this.opts,\n      ...opts,\n    } as BasetenTTSOptions;\n  }\n\n  /**\n   * Synthesize speech for a given piece of text.  Returns a `ChunkedStream`\n   * which will asynchronously fetch audio from Baseten and push frames into\n   * LiveKit's playback pipeline.  If you need to cancel synthesis you can\n   * call {@link ChunkedStream.stop} on the returned object.\n   */\n  synthesize(\n    text: string,\n    connOptions?: APIConnectOptions,\n    abortSignal?: AbortSignal,\n  ): ChunkedStream {\n    return new ChunkedStream(this, text, this.opts, connOptions, abortSignal);\n  }\n\n  stream(): tts.SynthesizeStream {\n    throw new Error('Streaming is not supported on Baseten TTS');\n  }\n\n  async close(): Promise<void> {\n    this.abortController.abort();\n  }\n}\n\n/**\n * Internal helper that performs the actual HTTP request and converts the\n * response into audio frames.  It inherits from `tts.ChunkedStream` to\n * integrate with LiveKit's event and cancellation framework.\n *\n * This implementation streams audio chunks as they arrive from the Baseten\n * model endpoint, processing them incrementally instead of waiting for the\n * complete response.\n */\nexport class ChunkedStream extends tts.ChunkedStream {\n  label = 'baseten.ChunkedStream';\n  private readonly opts: BasetenTTSOptions;\n\n  constructor(\n    tts: TTS,\n    text: string,\n    opts: BasetenTTSOptions,\n    connOptions?: APIConnectOptions,\n    abortSignal?: AbortSignal,\n  ) {\n    super(text, tts, connOptions, abortSignal);\n    this.opts = opts;\n  }\n\n  /**\n   * Execute the synthesis request.  This method is automatically invoked\n   * by the base class when the stream starts.  It performs a POST request\n   * to the configured `modelEndpoint` with the input text and optional\n   * parameters.  Audio chunks are streamed as they arrive and transformed\n   * into a sequence of `AudioFrame` objects that are enqueued immediately\n   * for playback.\n   */\n  protected async run() {\n    const { apiKey, modelEndpoint, voice, language, temperature, maxTokens } = this.opts;\n    const payload: Record<string, unknown> = {\n      prompt: this.inputText,\n    };\n    if (voice) payload.voice = voice;\n    if (language) payload.language = language;\n    if (temperature !== undefined) payload.temperature = temperature;\n    if (maxTokens !== undefined) payload.max_tokens = maxTokens;\n\n    const headers: Record<string, string> = {\n      Authorization: `Api-Key ${apiKey}`,\n      'Content-Type': 'application/json',\n    };\n\n    const response = await fetch(modelEndpoint, {\n      method: 'POST',\n      headers,\n      body: JSON.stringify(payload),\n      signal: this.abortSignal,\n    });\n\n    if (!response.ok) {\n      let errText: string;\n      try {\n        errText = await response.text();\n      } catch {\n        errText = response.statusText;\n      }\n      throw new Error(`Baseten TTS request failed: ${response.status} ${errText}`);\n    }\n\n    // Stream the response body as chunks arrive\n    if (!response.body) {\n      throw new Error('Response body is not available for streaming');\n    }\n\n    const requestId = shortuuid();\n    const audioByteStream = new AudioByteStream(24000, 1);\n    const reader = response.body.getReader();\n\n    try {\n      let lastFrame: AudioFrame | undefined;\n      const sendLastFrame = (segmentId: string, final: boolean) => {\n        if (lastFrame) {\n          this.queue.put({ requestId, segmentId, frame: lastFrame, final });\n          lastFrame = undefined;\n        }\n      };\n\n      // waitForAbort internally sets up an abort listener on the abort signal\n      // we need to put it outside loop to avoid constant re-registration of the listener\n      const abortPromise = waitForAbort(this.abortSignal);\n\n      while (!this.abortSignal.aborted) {\n        const result = await Promise.race([reader.read(), abortPromise]);\n\n        if (result === undefined) break; // aborted\n\n        const { done, value } = result;\n\n        if (done) {\n          break;\n        }\n\n        // Process the chunk and convert to audio frames\n        // Convert Uint8Array to ArrayBuffer for AudioByteStream\n        const frames = audioByteStream.write(value.buffer);\n\n        for (const frame of frames) {\n          sendLastFrame(requestId, false);\n          lastFrame = frame;\n        }\n      }\n\n      // Send the final frame\n      sendLastFrame(requestId, true);\n    } finally {\n      reader.releaseLock();\n      this.queue.close();\n    }\n  }\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAGA,oBAMO;AAIP,MAAM,oBAAgD;AAAA,EACpD,OAAO;AAAA,EACP,UAAU;AAAA,EACV,aAAa;AACf;AAKO,MAAM,YAAY,kBAAI,IAAI;AAAA,EACvB;AAAA,EACR,QAAQ;AAAA,EACA,kBAAkB,IAAI,gBAAgB;AAAA,EAC9C,YAAY,OAAmC,CAAC,GAAG;AAMjD,UAAM,MAAO,GAAG,EAAE,WAAW,MAAM,CAAC;AAGpC,UAAM,SAAS,KAAK,UAAU,QAAQ,IAAI;AAC1C,UAAM,gBAAgB,KAAK,iBAAiB,QAAQ,IAAI;AAExD,QAAI,CAAC,QAAQ;AACX,YAAM,IAAI;AAAA,QACR;AAAA,MACF;AAAA,IACF;AACA,QAAI,CAAC,eAAe;AAClB,YAAM,IAAI;AAAA,QACR;AAAA,MACF;AAAA,IACF;AAEA,SAAK,OAAO;AAAA,MACV,GAAG;AAAA,MACH,GAAG;AAAA,MACH;AAAA,MACA;AAAA,IACF;AAAA,EACF;AAAA,EAEA,cAAc,MAAoE;AAChF,SAAK,OAAO;AAAA,MACV,GAAG,KAAK;AAAA,MACR,GAAG;AAAA,IACL;AAAA,EACF;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAQA,WACE,MACA,aACA,aACe;AACf,WAAO,IAAI,cAAc,MAAM,MAAM,KAAK,MAAM,aAAa,WAAW;AAAA,EAC1E;AAAA,EAEA,SAA+B;AAC7B,UAAM,IAAI,MAAM,2CAA2C;AAAA,EAC7D;AAAA,EAEA,MAAM,QAAuB;AAC3B,SAAK,gBAAgB,MAAM;AAAA,EAC7B;AACF;AAWO,MAAM,sBAAsB,kBAAI,cAAc;AAAA,EACnD,QAAQ;AAAA,EACS;AAAA,EAEjB,YACEA,MACA,MACA,MACA,aACA,aACA;AACA,UAAM,MAAMA,MAAK,aAAa,WAAW;AACzC,SAAK,OAAO;AAAA,EACd;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAUA,MAAgB,MAAM;AACpB,UAAM,EAAE,QAAQ,eAAe,OAAO,UAAU,aAAa,UAAU,IAAI,KAAK;AAChF,UAAM,UAAmC;AAAA,MACvC,QAAQ,KAAK;AAAA,IACf;AACA,QAAI,MAAO,SAAQ,QAAQ;AAC3B,QAAI,SAAU,SAAQ,WAAW;AACjC,QAAI,gBAAgB,OAAW,SAAQ,cAAc;AACrD,QAAI,cAAc,OAAW,SAAQ,aAAa;AAElD,UAAM,UAAkC;AAAA,MACtC,eAAe,WAAW,MAAM;AAAA,MAChC,gBAAgB;AAAA,IAClB;AAEA,UAAM,WAAW,MAAM,MAAM,eAAe;AAAA,MAC1C,QAAQ;AAAA,MACR;AAAA,MACA,MAAM,KAAK,UAAU,OAAO;AAAA,MAC5B,QAAQ,KAAK;AAAA,IACf,CAAC;AAED,QAAI,CAAC,SAAS,IAAI;AAChB,UAAI;AACJ,UAAI;AACF,kBAAU,MAAM,SAAS,KAAK;AAAA,MAChC,QAAQ;AACN,kBAAU,SAAS;AAAA,MACrB;AACA,YAAM,IAAI,MAAM,+BAA+B,SAAS,MAAM,IAAI,OAAO,EAAE;AAAA,IAC7E;AAGA,QAAI,CAAC,SAAS,MAAM;AAClB,YAAM,IAAI,MAAM,8CAA8C;AAAA,IAChE;AAEA,UAAM,gBAAY,yBAAU;AAC5B,UAAM,kBAAkB,IAAI,8BAAgB,MAAO,CAAC;AACpD,UAAM,SAAS,SAAS,KAAK,UAAU;AAEvC,QAAI;AACF,UAAI;AACJ,YAAM,gBAAgB,CAAC,WAAmB,UAAmB;AAC3D,YAAI,WAAW;AACb,eAAK,MAAM,IAAI,EAAE,WAAW,WAAW,OAAO,WAAW,MAAM,CAAC;AAChE,sBAAY;AAAA,QACd;AAAA,MACF;AAIA,YAAM,mBAAe,4BAAa,KAAK,WAAW;AAElD,aAAO,CAAC,KAAK,YAAY,SAAS;AAChC,cAAM,SAAS,MAAM,QAAQ,KAAK,CAAC,OAAO,KAAK,GAAG,YAAY,CAAC;AAE/D,YAAI,WAAW,OAAW;AAE1B,cAAM,EAAE,MAAM,MAAM,IAAI;AAExB,YAAI,MAAM;AACR;AAAA,QACF;AAIA,cAAM,SAAS,gBAAgB,MAAM,MAAM,MAAM;AAEjD,mBAAW,SAAS,QAAQ;AAC1B,wBAAc,WAAW,KAAK;AAC9B,sBAAY;AAAA,QACd;AAAA,MACF;AAGA,oBAAc,WAAW,IAAI;AAAA,IAC/B,UAAE;AACA,aAAO,YAAY;AACnB,WAAK,MAAM,MAAM;AAAA,IACnB;AAAA,EACF;AACF;","names":["tts"]}

package/dist/tts.d.cts

@@ -0,0 +1,45 @@
+import { type APIConnectOptions, tts } from '@livekit/agents';
+import type { BasetenTTSOptions } from './types.js';
+/**
+ * Baseten TTS implementation (streaming, 24kHz mono)
+ */
+export declare class TTS extends tts.TTS {
+    private opts;
+    label: string;
+    private abortController;
+    constructor(opts?: Partial<BasetenTTSOptions>);
+    updateOptions(opts: Partial<Omit<BasetenTTSOptions, 'apiKey' | 'modelEndpoint'>>): void;
+    /**
+     * Synthesize speech for a given piece of text.  Returns a `ChunkedStream`
+     * which will asynchronously fetch audio from Baseten and push frames into
+     * LiveKit's playback pipeline.  If you need to cancel synthesis you can
+     * call {@link ChunkedStream.stop} on the returned object.
+     */
+    synthesize(text: string, connOptions?: APIConnectOptions, abortSignal?: AbortSignal): ChunkedStream;
+    stream(): tts.SynthesizeStream;
+    close(): Promise<void>;
+}
+/**
+ * Internal helper that performs the actual HTTP request and converts the
+ * response into audio frames.  It inherits from `tts.ChunkedStream` to
+ * integrate with LiveKit's event and cancellation framework.
+ *
+ * This implementation streams audio chunks as they arrive from the Baseten
+ * model endpoint, processing them incrementally instead of waiting for the
+ * complete response.
+ */
+export declare class ChunkedStream extends tts.ChunkedStream {
+    label: string;
+    private readonly opts;
+    constructor(tts: TTS, text: string, opts: BasetenTTSOptions, connOptions?: APIConnectOptions, abortSignal?: AbortSignal);
+    /**
+     * Execute the synthesis request.  This method is automatically invoked
+     * by the base class when the stream starts.  It performs a POST request
+     * to the configured `modelEndpoint` with the input text and optional
+     * parameters.  Audio chunks are streamed as they arrive and transformed
+     * into a sequence of `AudioFrame` objects that are enqueued immediately
+     * for playback.
+     */
+    protected run(): Promise<void>;
+}
+//# sourceMappingURL=tts.d.ts.map

package/dist/tts.d.ts

@@ -0,0 +1,45 @@
+import { type APIConnectOptions, tts } from '@livekit/agents';
+import type { BasetenTTSOptions } from './types.js';
+/**
+ * Baseten TTS implementation (streaming, 24kHz mono)
+ */
+export declare class TTS extends tts.TTS {
+    private opts;
+    label: string;
+    private abortController;
+    constructor(opts?: Partial<BasetenTTSOptions>);
+    updateOptions(opts: Partial<Omit<BasetenTTSOptions, 'apiKey' | 'modelEndpoint'>>): void;
+    /**
+     * Synthesize speech for a given piece of text.  Returns a `ChunkedStream`
+     * which will asynchronously fetch audio from Baseten and push frames into
+     * LiveKit's playback pipeline.  If you need to cancel synthesis you can
+     * call {@link ChunkedStream.stop} on the returned object.
+     */
+    synthesize(text: string, connOptions?: APIConnectOptions, abortSignal?: AbortSignal): ChunkedStream;
+    stream(): tts.SynthesizeStream;
+    close(): Promise<void>;
+}
+/**
+ * Internal helper that performs the actual HTTP request and converts the
+ * response into audio frames.  It inherits from `tts.ChunkedStream` to
+ * integrate with LiveKit's event and cancellation framework.
+ *
+ * This implementation streams audio chunks as they arrive from the Baseten
+ * model endpoint, processing them incrementally instead of waiting for the
+ * complete response.
+ */
+export declare class ChunkedStream extends tts.ChunkedStream {
+    label: string;
+    private readonly opts;
+    constructor(tts: TTS, text: string, opts: BasetenTTSOptions, connOptions?: APIConnectOptions, abortSignal?: AbortSignal);
+    /**
+     * Execute the synthesis request.  This method is automatically invoked
+     * by the base class when the stream starts.  It performs a POST request
+     * to the configured `modelEndpoint` with the input text and optional
+     * parameters.  Audio chunks are streamed as they arrive and transformed
+     * into a sequence of `AudioFrame` objects that are enqueued immediately
+     * for playback.
+     */
+    protected run(): Promise<void>;
+}
+//# sourceMappingURL=tts.d.ts.map

package/dist/tts.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"tts.d.ts","sourceRoot":"","sources":["../src/tts.ts"],"names":[],"mappings":"AAGA,OAAO,EACL,KAAK,iBAAiB,EAGtB,GAAG,EAEJ,MAAM,iBAAiB,CAAC;AAEzB,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,YAAY,CAAC;AAQpD;;GAEG;AACH,qBAAa,GAAI,SAAQ,GAAG,CAAC,GAAG;IAC9B,OAAO,CAAC,IAAI,CAAoB;IAChC,KAAK,SAAiB;IACtB,OAAO,CAAC,eAAe,CAAyB;gBACpC,IAAI,GAAE,OAAO,CAAC,iBAAiB,CAAM;IA+BjD,aAAa,CAAC,IAAI,EAAE,OAAO,CAAC,IAAI,CAAC,iBAAiB,EAAE,QAAQ,GAAG,eAAe,CAAC,CAAC;IAOhF;;;;;OAKG;IACH,UAAU,CACR,IAAI,EAAE,MAAM,EACZ,WAAW,CAAC,EAAE,iBAAiB,EAC/B,WAAW,CAAC,EAAE,WAAW,GACxB,aAAa;IAIhB,MAAM,IAAI,GAAG,CAAC,gBAAgB;IAIxB,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC;CAG7B;AAED;;;;;;;;GAQG;AACH,qBAAa,aAAc,SAAQ,GAAG,CAAC,aAAa;IAClD,KAAK,SAA2B;IAChC,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAoB;gBAGvC,GAAG,EAAE,GAAG,EACR,IAAI,EAAE,MAAM,EACZ,IAAI,EAAE,iBAAiB,EACvB,WAAW,CAAC,EAAE,iBAAiB,EAC/B,WAAW,CAAC,EAAE,WAAW;IAM3B;;;;;;;OAOG;cACa,GAAG;CAkFpB"}