Function introduction
Which voices does speech synthesis support?
For the voices supported by CosyVoice-V2, see the CosyVoice voice list (CosyVoice-V2) in Real-time speech synthesis - CosyVoice/Sambert.
For the voices supported by Sambert, see the Sambert voice list in Real-time speech synthesis - CosyVoice/Sambert.
For the voices supported by Qwen-TTS, see Real-time speech synthesis - Qwen.
Which on-device algorithms are supported? Are they just for wake words and instructions, or are there other capabilities?
The on-device software development kit (SDK) integrates voice wake-up, on-device Voice Activity Detection (VAD), Acoustic Echo Cancellation (AEC), and directional sound detection algorithms. For more information, see On-device algorithms.
In real-time interactive voice response scenarios, what are the differences between the push2talk, tap2talk, and duplex upstream modes?
Push2Talk mode: Press and hold the button to start recording. Release the button to stop recording.
Tap2Talk mode: Click once to start recording. Click again to stop recording.
Duplex mode: Supports simultaneous bidirectional communication. This enables real-time conversation where you can listen and speak at the same time.
Which languages does speech translation support?
For more information, see Speech translation.
Configuration and development
What are the format requirements for audio data collection and playback?
For more information, see Audio collection and playback instructions.
When calling an API, do I need to generate a new task_id each time?
When you start a new session, use a new task_id. The task_id serves as the session identity and can affect configuration or status determination. If you use the SDK to make calls, you do not need to manage this. The SDK automatically generates a new task_id for each new session.
How can I add special business logic to a prompt and pass it to the large model?
When you configure prompts, you can use the custom variables feature. For more information, see Prompts.
Which types of Alibaba Cloud Model Studio workspaces are supported?
Currently, only the main workspace is supported. To connect, use the workspaceId of your main workspace.
Deployment
Is deployment outside China supported?
The Alibaba Cloud Model Studio multimodal interaction development suite can be accessed and used from outside China. However, deployment on service nodes outside China is not yet supported. To find solutions for reducing network latency when accessing from outside China, contact our business team.
Troubleshooting
Why does the text-to-speech (TTS) playback include content that is not in the original text?
Check the text for special formats or characters, such as Markdown. The TTS model currently reads Markdown content aloud. You can constrain the model to not output content in Markdown format, or filter special characters from the output text.
The sample audio on the official website works, but I cannot get a recognition result when I use my own test audio. Why?
Check the audio file format:
- Ensure that the format of your test audio meets the input format requirements for speech recognition. For more information about the requirements, see Audio collection and playback instructions.
- Convert the test audio to an uncompressed WAV file with a 16 kHz sample rate, 16-bit audio bit depth, and a single sound channel (mono).