ChunkPlan
This is the chunk plan for the voice output. It controls how the model output is split into chunks
before being sent to the voice provider for generation.
Default true because voice generation sounds better with chunking (and reformatting them).
To send every token from the model output directly to the voice provider and rely on the voice provider's audio
generation logic, set enabled to false.
Properties
This determines whether the model output is preprocessed into chunks before being sent to the voice provider.
Default true because voice generation sounds better with chunking (and reformatting them).
To send every token from the model output directly to the voice provider and rely on the voice provider's audio
generation logic, set this to false.
If disabled, vapi-provided audio control tokens like <flush /> will not work.
This is the minimum number of characters before a chunk is created. The chunks that are sent to the voice provider
for the voice generation as the model tokens are streaming in. Defaults to 30.
Increasing this value might add latency as it waits for the model to output a full chunk before sending it to the
voice provider. On the other hand, increasing might be a good idea if you want to give voice provider bigger chunks,
so it can pronounce them better.
Decreasing this value might decrease latency but might also decrease quality if the voice provider struggles to
pronounce the text correctly.
These are the punctuations that are considered valid boundaries before a chunk is created. The chunks that are sent
to the voice provider for the voice generation as the model tokens are streaming in. Defaults are chosen differently
for each provider.
Constraining the delimiters might add latency as it waits for the model to output a full chunk before sending it to
the voice provider. On the other hand, constraining might be a good idea if you want to give voice provider longer
chunks, so it can sound less disjointed across chunks. Eg. ['.'].