Use simple api call, write binary response to tmp file and read back out

app.py

@@ -14,17 +14,20 @@ token_str = os.environ['DG_TOKEN']
 def tts_fn(text, prompt_audio, prompt_seconds, inference_steps, inference_temperature, pitch_steps):
     texts = [text]
     sr = prompt_audio[0]
-    prompt_audio = np.reshape(prompt_audio[1], (1, -1)).astype(np.float32, order='C') / 32768.0
+    prompt_audio = prompt_audio[1].astype(np.float32, order='C') / 32768.0
     byte_io = io.BytesIO(bytes())
     wavfile.write(byte_io, sr, prompt_audio)
     prompt_audio_bytes = byte_io.read()
-    prompt_audio = [base64.b64encode(prompt_audio_bytes).decode('utf-8')] * len(texts)
     params={'synthesize': 'true', 'pitch_steps': int(pitch_steps), 'soundstorm_steps': inference_steps, 'temperature': inference_temperature, 'prompt_seconds': prompt_seconds}
-    files=[('texts', ('texts', json.dumps(texts), 'application/json')), ('prompt_audio', ('prompt_audio', json.dumps(prompt_audio), 'application/json'))]
-    response = requests.post(base_url, files=files, params=params, headers={'Authorization': f'Token {token_str}'}).json()
+    response = requests.post(base_url, data=prompt_audio_bytes, params=params, headers={'Authorization': f'Token {token_str}'})
     try:
-        sample_rate = int(response['results'][0]['sample_rate'])
-        audio = (np.array(response['results'][0]['audio']).transpose() / 1.414 * 32767).astype(np.int16)
+        with open('result.wav', 'wb') as f:
+            for chunk in response.iter_content(chunk_size=1024):
+                if chunk: f.write(chunk)
+        sample_rate, audio = wavfile.read('result.wav')
+        print(audio.dtype)
+        audio = (audio / 1.414 * 32767).astype(np.int16)
+        print(audio.dtype)
     except Exception:
         print(response)
     return (sample_rate, audio)