Salut les francophone, vous avez vu cette update de cartesia : Hey @everyone ! We just released our continuations feature in our API. This allows you to condition generations on previous ones such that chunked generations sound like one contiguous generation - this is particularly useful for streaming input text (for example from an LLM) for conversational applications or to improve quality on long generations through chunking. Here’s an example of a long, chunked generation with continuations:

transcripts = [
    "Magnetic resonance imaging (MRI) is a cornerstone of modern healthcare.",
    "However, long scan times and the need for qualitative expert image analysis have curtailed its widespread use.",
    "Recent advances in machine learning (ML) have shown promise for reducing this burden,",
    "but there are several factors that have limited the development and deployment of these methods in practice.",
    "First, current ML methods require large amounts of labeled training data,",
    "which are often impossible to collect,",
    "and fail under data distribution shifts,",
    "which occur frequently during image acquisition (e.g., changes in signal-to-noise ratio and anatomy).",
]

If it's relevant for your use case, would love any feedback! Here’s the docs for it https://docs.cartesia.ai/api-reference/endpoints/stream-speech-websocket#input-streaming-with-contexts and it's also available in the latest client version. We're working on using this to natively support long generations through chunking via the API this week.