Python Workflow | Resemble

Upload a folder of recordings and transcripts, then build a production-ready voice.

Prerequisites

Python 3.8+
PRO plan (or higher) account and API token
Folder containing paired .wav and .txt files (clip-01.wav + clip-01.txt, etc.)

python3 -m venv venv
source venv/bin/activate
pip install resemble

1. Script scaffold

# resemble-clone-voice-recording/main.py
import argparse
import os
from resemble import Resemble
def initialize() -> None:
    api_key = os.environ.get("RESEMBLE_API_KEY")
    if not api_key:
        raise EnvironmentError("Set RESEMBLE_API_KEY before running.")
    Resemble.api_key(api_key)
def parse_args():
    parser = argparse.ArgumentParser(
        description="Create a voice from local recordings"
    )
    parser.add_argument("--name", required=True)
    parser.add_argument("--recordings", required=True)
    return parser.parse_args()

2. Create the voice

def create_voice(name: str) -> str:
    print(f"Creating voice {name}...")
    response = Resemble.v2.voices.create(name=name)
    if not response["success"]:
        raise RuntimeError(response)
    voice = response["item"]
    print(f"Voice UUID: {voice['uuid']} (status: {voice['status']})")
    return voice["uuid"]

3. Prepare recordings

import os
def read_folder(folder_path: str):
    entries = []
    for filename in os.listdir(folder_path):
        if not filename.endswith(".wav"):
            continue
        transcript = filename.replace(".wav", ".txt")
        transcript_path = os.path.join(folder_path, transcript)
        if not os.path.exists(transcript_path):
            print(f"Skipping {filename}; missing transcript.")
            continue
        with open(transcript_path, "r", encoding="utf-8") as handle:
            text = handle.read()
        entries.append(
            {
                "file_path": os.path.join(folder_path, filename),
                "name": transcript,
                "text": text,
            }
        )
    return entries

Aim for at least 20 clean samples (1–12 seconds, no silence). Longer files are ignored during training.

4. Upload recordings

def upload_recordings(voice_uuid: str, folder_path: str) -> None:
    recordings = read_folder(folder_path)
    successes = 0
    for recording in recordings:
        print(f"Uploading {recording['name']}...")
        with open(recording["file_path"], "rb") as audio_file:
            response = Resemble.v2.recordings.create(
                voice_uuid,
                audio_file,
                recording["name"],
                recording["text"],
                is_active=True,
                emotion="neutral",
            )
        if response["success"]:
            successes += 1
        else:
            print(f"Failed to upload {recording['name']}")
            print(response)
    print(f"Uploaded {successes}/{len(recordings)} recordings")

5. Trigger the build

def trigger_voice_build(voice_uuid: str) -> None:
    response = Resemble.v2.voices.build(uuid=voice_uuid)
    if not response["success"]:
        raise RuntimeError(response)
    print("Build request submitted. Monitor progress via the API or dashboard.")

6. Wire everything together

def main():
    args = parse_args()
    initialize()
    voice_uuid = create_voice(args.name)
    upload_recordings(voice_uuid, args.recordings)
    trigger_voice_build(voice_uuid)
if __name__ == "__main__":
    main()

RESEMBLE_API_KEY=... python main.py --name "Support Voice" --recordings ./example-data

The script prints upload progress and starts training. Use List Voices or the dashboard to monitor build status until the voice is ready.