Voice Cloning

Create a voice

1. Log on to the Intelligent Media Services console.

2. In the left-side navigation pane, choose Real-time Conversational AI > Voice Cloning.

3. Click Create Voice. In the upper-right corner, the page displays your current voice count and quota limit (30 voices).

4. On the Create Voice page, configure the following parameters:

   • Voice Name: Enter a custom name for the voice. This helps you identify and select it in workflows.

   • Original Audio: Select the audio source method:

     • Upload an audio file first.: Select and upload an audio file from your local machine that meets the requirements.

     • Record: Click the record button to record audio from your microphone. The recording must be at least 20 seconds long.

   • Cloning Model: Select the model to use for voice cloning. The following models are supported:

     • CosyVoice series: cosyvoice-v2, cosyvoice-v3-flash, and cosyvoice-v3-plus

     For more information about these models and their capabilities, see Real-time speech synthesis - CosyVoice/Sambert and Real-time speech synthesis - Qwen.

5. Click Submit. The system starts processing the voice cloning request and initiates a review.

6. On the Voice Cloning list page, check the voice status:

   • Reviewing: The voice is under review and cannot be used yet.

   • Approved: The voice is ready to use. Click the Audition button to listen to a synthesized sample.

   • Review Failed: The voice did not pass the review and cannot be used. Check if your audio content complies with the terms of service.

7. Once the voice is approved, find its voice ID on the list page. You will need this ID to configure the tts node in your workflow.

List page features:

• Search for voices by name or ID.

• Each account has a quota limit of 30 voices. To create a new voice after reaching this limit, you must first delete an existing one.

Upload audio to OSS

After preparing your audio files, upload them to an OSS bucket. For instructions, see Procedures.

Call the API to get a voice ID

The following sample code demonstrates how to call the Model Studio API to generate a voice ID from an audio file stored in OSS.

import os
import dashscope
from dashscope.audio.tts_v2 import VoiceEnrollmentService, SpeechSynthesizer

# Obtain the API key from an environment variable to avoid hardcoding.
dashscope.api_key = os.getenv('DASHSCOPE_API_KEY')
url = "https://your-audio-file-url"  # Replace this with the public URL of your audio file in OSS.
prefix = 'prefix'  # Define a custom prefix to identify the voice source.
target_model = "cosyvoice-v2"  # Select a cloning model.

# Create an instance of the voice enrollment service.
service = VoiceEnrollmentService()

# Call the create_voice method to clone the voice and generate a voice_id.
voice_id = service.create_voice(target_model=target_model, prefix=prefix, url=url)
print(f"your voice id is {voice_id}")
# Sample output: cosyvoice-prefix-xxxxx

After the call completes, save the returned voice_id string, which is your voice ID. Voices created via the API also appear on the Voice Cloning list page, where you can view their status and manage them.