🎵 PicoAudio2 Online Inference

TCC (Temporal Coarse Caption)
A brief text description for the overall audio scene.

Example: a dog barks

TDC (Temporal Detailed Caption)
Event descriptions with timestamps. It allows precise temporal control over when events happen in the generated audio.

Example: See the format below.

💡 JSON Format:

• Each event description is a key in the JSON object
• The value is an array of [start, end] timestamp pairs (in seconds)

Example:

{
    "a dog barks": [
        [3.0, 4.0],
        [6.0, 7.0]
    ],
    "a man speaks": [
        [5.0, 6.0]
    ]
}

It means the event a dog barks happens from 3.0 to 4.0 seconds, and the event a man speaks happens from 5.0 to 6.0 seconds.

1. TCC is required for audio generation.

2. TDC (JSON) is optional for precise temporal control of events.

3. Length (in seconds) is optional, but recommended for temporal control. Defaults to 10.0 seconds.

4. Enable Time Control: Tick to use TDC and length for precise event timing.

1. Currently events with overlapped timestamps will not get good results, so we recommend generating audio without temporal overlaps between events.

2. If TDC format is incorrect or length is missing, the model will generate audio without precise temporal control.

3. For general audio generation without precise timing, you can leave the JSON empty.

4. You may leave TDC blank to let the LLM generate timestamps automatically (subject to API quota).