`voxcpm2-continuation`

Continue a spoken prompt in the same voice.

Native waveform: Prompt encoded at 16 kHz. Generate 48 kHz mono.. Live price and availability are on GET /v1/models/voxcpm2-continuation.

text is the new words. reference_text is the transcript of the prompt clip. Both are required.
Credits are held on the new text, not the prompt length.

Parameters

Field	Type	Description
`params.cfg_value` Default 2	number	VoxCPM2 only. 1 to 3. Higher follows the text more tightly. Default 2.
`params.inference_timesteps` Default 10	integer	VoxCPM2 only. 4 to 30. More is usually cleaner and slower. Default 10.
`params.seed` Default 0	integer	VoxCPM2 only. Official generate seed. 0 is a new take each run. Same on speech, clone, design, and continuation. Does not change RTF.
`params.output_format` Default wav (flac for stems)	string	wav, flac, mp3, or model (the model native file).
`params.output_sample_rate` Default source	string or integer	source keeps the upload (or the model rate when there is no upload). model uses the model native rate. Or pass 8000, 16000, 22050, 24000, 32000, 44100, or 48000.
`params.output_channels` Default source	string	source, model, mono, or stereo. Applied after inference, not before.

Example

curl https://api.pascalai.co/v1/voice-continuation \
  -H "Authorization: Bearer $PASCAL_API_KEY" \
  -H "Idempotency-Key: $(uuidgen)" \
  -F model=voxcpm2-continuation \
  -F text="and then we kept walking toward the river." \
  -F reference_text="Hello, my name is Ada." \
  -F params='{"cfg_value":2,"output_format":"wav"}' \
  -F file=@prompt.wav

curl https://api.pascalai.co/v1/voice-continuation/$JOB_ID \
  -H "Authorization: Bearer $PASCAL_API_KEY"