pareta 1.1.0 → 1.2.0

package/dist/index.cjs

@@ -174,7 +174,7 @@ function parseSSE(reader, opts) {
 }
 
 // src/version.ts
-var VERSION = "1.1.0";
+var VERSION = "1.2.0";
 
 // src/money.ts
 var MICRO_PER_CENT = 10000n;
@@ -486,6 +486,53 @@ var Embeddings = class extends BaseModel {
     return (this.raw.data ?? []).length;
   }
 };
+function base64ToBytes(b64) {
+  if (typeof Buffer !== "undefined") return new Uint8Array(Buffer.from(b64, "base64"));
+  const bin = atob(b64);
+  const bytes = new Uint8Array(bin.length);
+  for (let i = 0; i < bin.length; i++) bytes[i] = bin.charCodeAt(i);
+  return bytes;
+}
+var Transcription = class extends BaseModel {
+  get text() {
+    return this.raw.text ?? null;
+  }
+  get language() {
+    return this.raw.language ?? null;
+  }
+  get durationS() {
+    return this.raw.duration_s ?? null;
+  }
+  toString() {
+    return this.text ?? "";
+  }
+};
+var Speech = class extends BaseModel {
+  /** The synthesized audio, base64-decoded to raw bytes. */
+  get audio() {
+    const b64 = this.raw.audio_base64 || "";
+    return b64 ? base64ToBytes(b64) : new Uint8Array(0);
+  }
+  get audioBase64() {
+    return this.raw.audio_base64 ?? null;
+  }
+  get sampleRate() {
+    return this.raw.sample_rate ?? null;
+  }
+  get durationS() {
+    return this.raw.duration_s ?? null;
+  }
+  /** Container/codec of the returned audio (e.g. "wav"). */
+  get format() {
+    return this.raw.format ?? null;
+  }
+  /** Write the decoded audio to `path` (Node only — lazy node:fs). */
+  async save(path) {
+    const { writeFile } = await import('fs/promises');
+    await writeFile(path, this.audio);
+    return this;
+  }
+};
 
 // src/resources/chat.ts
 var PATH = "/v1/chat/completions";
@@ -788,6 +835,60 @@ var Auto = class {
   }
 };
 
+// src/resources/audio.ts
+var BASE3 = "/v1/audio";
+function bytesToBase64(bytes) {
+  if (typeof Buffer !== "undefined") return Buffer.from(bytes).toString("base64");
+  let bin = "";
+  const CHUNK = 32768;
+  for (let i = 0; i < bytes.length; i += CHUNK) {
+    bin += String.fromCharCode(...bytes.subarray(i, i + CHUNK));
+  }
+  return btoa(bin);
+}
+async function toBase64(audio) {
+  if (typeof audio === "string") {
+    const { readFile } = await import('fs/promises');
+    return bytesToBase64(new Uint8Array(await readFile(audio)));
+  }
+  if (typeof audio === "object" && audio !== null && "base64" in audio) {
+    if (!audio.base64 || !audio.base64.trim()) throw new ParetaError("audio.base64 is empty");
+    return audio.base64;
+  }
+  if (audio instanceof Blob) return bytesToBase64(new Uint8Array(await audio.arrayBuffer()));
+  const bytes = audio instanceof Uint8Array ? audio : new Uint8Array(audio);
+  if (bytes.byteLength === 0) throw new ParetaError("audio is empty");
+  return bytesToBase64(bytes);
+}
+var Audio = class {
+  constructor(client) {
+    this.client = client;
+  }
+  client;
+  /** Speech-to-text (the `asr` lane). `audio` is a file path, raw bytes, a
+   * Blob, or `{ base64 }`. Metered per minute of input audio. */
+  async transcriptions(audio, opts = {}) {
+    const body = { audio_base64: await toBase64(audio) };
+    if (opts.language) body.language = opts.language;
+    return this.client.request("POST", `${BASE3}/transcriptions`, {
+      body,
+      cast: (raw) => new Transcription(raw)
+    });
+  }
+  /** Text-to-speech (the `tts` lane). Returns a `Speech` whose `.audio` is
+   * decoded bytes (`.save(path)` writes a file in Node). Metered per minute
+   * of output audio. */
+  speech(text, opts = {}) {
+    if (!text || !text.trim()) throw new ParetaError("text is required");
+    const body = { text };
+    if (opts.voice) body.voice = opts.voice;
+    return this.client.request("POST", `${BASE3}/speech`, {
+      body,
+      cast: (raw) => new Speech(raw)
+    });
+  }
+};
+
 // src/resources/retrieval.ts
 function makeRerank(client) {
   return (query, documents, opts = {}) => {
@@ -848,6 +949,8 @@ var Pareta = class _Pareta {
   tasks;
   evals;
   auto;
+  /** The Speech lanes — `pa.audio.transcriptions(...)` (ASR) + `pa.audio.speech(...)` (TTS). */
+  audio;
   /** Document reranking (the Retrieval precision lane) — `pa.rerank(query, docs, {topN})`. */
   rerank;
   /** Text embeddings (the Retrieval recall lane) — `pa.embeddings(input, {inputType})`. */
@@ -872,6 +975,7 @@ var Pareta = class _Pareta {
     this.tasks = new Tasks(this);
     this.evals = new Evals(this);
     this.auto = new Auto(this);
+    this.audio = new Audio(this);
     this.rerank = makeRerank(this);
     this.embeddings = makeEmbeddings(this);
   }
@@ -1057,6 +1161,7 @@ var Pareta = class _Pareta {
 exports.APIConnectionError = APIConnectionError;
 exports.APIStatusError = APIStatusError;
 exports.APITimeoutError = APITimeoutError;
+exports.Audio = Audio;
 exports.AuthenticationError = AuthenticationError;
 exports.BadRequestError = BadRequestError;
 exports.BaseModel = BaseModel;
@@ -1084,9 +1189,11 @@ exports.PermissionDeniedError = PermissionDeniedError;
 exports.RateLimitError = RateLimitError;
 exports.Rerank = Rerank;
 exports.RerankResult = RerankResult;
+exports.Speech = Speech;
 exports.Task = Task;
 exports.TaskMatch = TaskMatch;
 exports.TaskMatchCandidate = TaskMatchCandidate;
+exports.Transcription = Transcription;
 exports.Usage = Usage;
 exports.VERSION = VERSION;
 //# sourceMappingURL=index.cjs.map