diff --git a/frontend/server/utils/whisper.ts b/frontend/server/utils/whisper.ts
index ffbb903..2a3dd61 100644
--- a/frontend/server/utils/whisper.ts
+++ b/frontend/server/utils/whisper.ts
@@ -3,11 +3,9 @@ type WhisperTranscribeInput = {
   sampleRate: number;
   language?: string;
 };
-let whisperPipelinePromise: Promise<any> | null = null;
-let transformersPromise: Promise<any> | null = null;
 
-function getWhisperModelId() {
-  return (process.env.CF_WHISPER_MODEL ?? "Xenova/whisper-small").trim() || "Xenova/whisper-small";
+function getWhisperUrl() {
+  return (process.env.WHISPER_URL ?? "http://whisper:9000").replace(/\/+$/, "");
 }
 
 function getWhisperLanguage() {
@@ -15,39 +13,57 @@ function getWhisperLanguage() {
   return value || "ru";
 }
 
-async function getWhisperPipeline() {
-  if (!transformersPromise) {
-    transformersPromise = import("@xenova/transformers");
+function pcmToWav(samples: Float32Array, sampleRate: number): Buffer {
+  const numChannels = 1;
+  const bitsPerSample = 16;
+  const byteRate = sampleRate * numChannels * (bitsPerSample / 8);
+  const blockAlign = numChannels * (bitsPerSample / 8);
+  const dataSize = samples.length * (bitsPerSample / 8);
+  const headerSize = 44;
+  const buffer = Buffer.alloc(headerSize + dataSize);
+
+  // RIFF header
+  buffer.write("RIFF", 0);
+  buffer.writeUInt32LE(36 + dataSize, 4);
+  buffer.write("WAVE", 8);
+
+  // fmt chunk
+  buffer.write("fmt ", 12);
+  buffer.writeUInt32LE(16, 16);
+  buffer.writeUInt16LE(1, 20); // PCM
+  buffer.writeUInt16LE(numChannels, 22);
+  buffer.writeUInt32LE(sampleRate, 24);
+  buffer.writeUInt32LE(byteRate, 28);
+  buffer.writeUInt16LE(blockAlign, 32);
+  buffer.writeUInt16LE(bitsPerSample, 34);
+
+  // data chunk
+  buffer.write("data", 36);
+  buffer.writeUInt32LE(dataSize, 40);
+
+  for (let i = 0; i < samples.length; i += 1) {
+    const s = Math.max(-1, Math.min(1, samples[i] ?? 0));
+    const val = s < 0 ? s * 0x8000 : s * 0x7fff;
+    buffer.writeInt16LE(Math.round(val), headerSize + i * 2);
   }
 
-  const { env, pipeline } = await transformersPromise;
-
-  if (!whisperPipelinePromise) {
-    env.allowRemoteModels = true;
-    env.allowLocalModels = true;
-    env.cacheDir = "/app/.data/transformers";
-
-    const modelId = getWhisperModelId();
-    whisperPipelinePromise = pipeline("automatic-speech-recognition", modelId);
-  }
-
-  return whisperPipelinePromise;
+  return buffer;
 }
 
 export async function transcribeWithWhisper(input: WhisperTranscribeInput) {
-  const transcriber = (await getWhisperPipeline()) as any;
-  const result = await transcriber(
-    input.samples,
-    {
-      sampling_rate: input.sampleRate,
-      language: (input.language ?? getWhisperLanguage()) || "ru",
-      task: "transcribe",
-      chunk_length_s: 20,
-      stride_length_s: 5,
-      return_timestamps: false,
-    },
-  );
+  const wav = pcmToWav(input.samples, input.sampleRate);
+  const language = (input.language ?? getWhisperLanguage()) || "ru";
+  const url = `${getWhisperUrl()}/asr?task=transcribe&language=${language}&output=json`;
 
-  const text = String((result as any)?.text ?? "").trim();
-  return text;
+  const formData = new FormData();
+  formData.append("audio_file", new Blob([wav], { type: "audio/wav" }), "audio.wav");
+
+  const response = await fetch(url, { method: "POST", body: formData });
+  if (!response.ok) {
+    const detail = await response.text().catch(() => "");
+    throw new Error(`Whisper service error ${response.status}: ${detail}`);
+  }
+
+  const result = (await response.json()) as { text?: string };
+  return String(result?.text ?? "").trim();
 }