$ cat node-template.py
Speech Creation
// Converts text to speech with support for 23 languages and optional voice cloning from a reference audio file using ChatterBox v2 via a native GPU service. Adjustable expressiveness and generation temperature. Outputs an MP3 audio file.
Process
Audio
template.py
1import os2import sys3import json4import subprocess5import time6import traceback78try:9 import requests10except ImportError:11 subprocess.check_call([sys.executable, "-m", "pip", "install", "requests"])12 import requests1314NATIVE_SPEECH_CREATION_SERVICE_URL = os.getenv(15 "NATIVE_SPEECH_CREATION_SERVICE_URL", "http://native-speech-creation-service:8108"16)17_EMBLEMA_VERSION = os.getenv("EMBLEMA_VERSION", "dev")18NATIVE_SPEECH_CREATION_SERVICE_IMAGE = os.getenv(19 "NATIVE_SPEECH_CREATION_SERVICE_IMAGE",20 f"emblema/native-speech-creation-service:{_EMBLEMA_VERSION}",21)22HF_CACHE_HOST_PATH = os.getenv("HF_CACHE_HOST_PATH", "/root/.cache/huggingface")23CONTAINER_NAME = "native-speech-creation-service"24INPUT_DIR = "/data/input"25OUTPUT_DIR = "/data/output"262728def start_container():29 """Create and start native-speech-creation-service, removing any stale container first."""30 subprocess.run(31 ["docker", "rm", "-f", CONTAINER_NAME],32 capture_output=True, text=True33 )3435 hf_token = os.getenv("HUGGINGFACE_TOKEN", "")36 print(f"Creating container {CONTAINER_NAME}...", file=sys.stderr)37 run_cmd = [38 "docker", "run", "-d",39 "--name", CONTAINER_NAME,40 "--network", "emblema",41 "--gpus", "all",42 "-e", "PORT=8108",43 "-e", "DEVICE=cuda",44 "-e", f"HF_TOKEN={hf_token}",45 "-v", f"{HF_CACHE_HOST_PATH}:/root/.cache/huggingface",46 NATIVE_SPEECH_CREATION_SERVICE_IMAGE,47 ]48 result = subprocess.run(run_cmd, capture_output=True, text=True)49 if result.returncode != 0:50 print(f"docker run failed (exit {result.returncode}): {result.stderr}", file=sys.stderr)51 raise RuntimeError(f"Failed to start container: {result.stderr}")5253 # Poll health endpoint54 timeout = 18055 interval = 356 elapsed = 057 health_url = f"{NATIVE_SPEECH_CREATION_SERVICE_URL}/health"58 while elapsed < timeout:59 try:60 r = requests.get(health_url, timeout=5)61 if r.status_code == 200:62 print(f"Container healthy (waited {elapsed}s).", file=sys.stderr)63 return64 except requests.ConnectionError:65 pass66 time.sleep(interval)67 elapsed += interval6869 raise RuntimeError(f"Container did not become healthy within {timeout}s")707172def stop_container():73 """Remove the container."""74 try:75 subprocess.run(76 ["docker", "rm", "-f", CONTAINER_NAME],77 capture_output=True, text=True, timeout=3078 )79 print(f"Container {CONTAINER_NAME} removed.", file=sys.stderr)80 except Exception as e:81 print(f"Warning: failed to remove container: {e}", file=sys.stderr)828384def main():85 try:86 input_json = sys.stdin.read()87 execution_input = json.loads(input_json)88 inputs = execution_input.get("inputs", {})8990 prompt = inputs.get("prompt", "")91 voice_reference = inputs.get("voice_reference", "")92 language = inputs.get("language", "Italian")93 exaggeration = float(inputs.get("exaggeration", 0.8))94 temperature = float(inputs.get("temperature", 0.8))9596 if not prompt:97 raise ValueError("Prompt input is required")98 if not (0.0 <= exaggeration <= 1.0):99 raise ValueError(f"Exaggeration must be between 0.0 and 1.0, got {exaggeration}")100 if not (0.0 <= temperature <= 1.5):101 raise ValueError(f"Temperature must be between 0.0 and 1.5, got {temperature}")102103 os.makedirs(OUTPUT_DIR, exist_ok=True)104105 # Start the container106 start_container()107108 try:109 # Build multipart form data110 form_data = {111 "text": (None, prompt),112 "language": (None, language),113 "exaggeration": (None, str(exaggeration)),114 "temperature": (None, str(temperature)),115 }116117 # Optionally attach voice reference file118 if voice_reference:119 voice_path = os.path.join(INPUT_DIR, voice_reference)120 if not os.path.exists(voice_path):121 raise FileNotFoundError(f"Voice reference not found: {voice_path}")122 form_data["voice_reference"] = (123 os.path.basename(voice_path),124 open(voice_path, "rb"),125 "audio/mpeg",126 )127128 resp = requests.post(129 f"{NATIVE_SPEECH_CREATION_SERVICE_URL}/generate",130 files=form_data,131 timeout=600,132 )133134 if resp.status_code != 200:135 try:136 error_detail = resp.json()137 except Exception:138 error_detail = resp.text139 raise RuntimeError(140 f"Speech creation service returned {resp.status_code}: {error_detail}"141 )142143 # Save result144 out_filename = "generated_speech.mp3"145 out_path = os.path.join(OUTPUT_DIR, out_filename)146 with open(out_path, "wb") as f:147 f.write(resp.content)148149 inference_time = resp.headers.get("X-Inference-Time-Ms", "unknown")150 print(151 f"Speech generated: time={inference_time}ms, language={language}, "152 f"exaggeration={exaggeration}, temperature={temperature}, "153 f"voice_cloning={bool(voice_reference)}",154 file=sys.stderr,155 )156157 # Flat output -- keys match OUTPUT_SCHEMA158 output = {159 "audio": out_filename,160 }161 print(json.dumps(output, indent=2))162163 finally:164 stop_container()165166 except Exception as e:167 error_output = {168 "error": str(e),169 "errorType": type(e).__name__,170 "traceback": traceback.format_exc(),171 }172 print(json.dumps(error_output), file=sys.stderr)173 sys.exit(1)174175176if __name__ == "__main__":177 main()