TTS volume 0.5 to 2: set narration gain before mixing with music

The knobs

The Sume TTS 1.0 request has a generation_config object. In the API reference, volume ranges from 0.5 to 2, speed from 0.6 to 1.5, and emotion is a string of 1 to 64 characters. A separate deprecated speed field takes slow, normal or fast, so use the numeric one.

Why set it at generation time

Level matching is easier before the mix than after it. If narration is too quiet against a music bed, you can raise volume at generation and avoid applying gain later on a clip that might already sit near full scale.

I have not verified how the volume scale maps to decibels, so measure one clip rather than assuming 2 means a particular gain.

A request

Keep wav through the mix so you do not re-encode. See wav vs mp3 for the reason.

{
  "transcript": "Welcome back. Here is this week's update.",
  "generation_config": { "volume": 1.2, "speed": 1.0 },
  "output_format": { "container": "wav" }
}

A mixing routine

• Render a 10-second sample at volume 1.0 and listen against your music bed.
• If the voice is quiet, try 1.2 to 1.5; if it clips, step down.
• Keep the same volume across every line of a script so levels do not jump.
• Render the final lines and join them with Timeline audio.

Speed and level together

Speed and volume are separate fields, and the ranges are different: 0.6 to 1.5 for speed, 0.5 to 2 for volume. A slower read is longer, so it uses more of the 1,200-second limit, while a louder read does not change duration.

Pick speed first for pacing, then volume for level, and render a sample of both before committing to a long script.

Takeaway

Set volume between 0.5 and 2 for first-pass level, keep to one value per project and measure rather than guess. Music from the router comes at its own level, so balance the two on a short test before you render a long read.