Agreed. The API and voice library are solid, but the post-generation workflow is missing a core feature.
Every time I generate dialogue, I'm left with:
* Separate audio files for each character/line
* No built-in tool to adjust relative volume, add pauses, or crossfade
* Have to drop everything into Audacity or Descript, which defeats the purpose of an "all-in-one" platform
Example: Trying to create a simple back-and-forth conversation.
```json
{
"lines": [
{"voice": "John", "text": "Are you sure about that?"},
{"voice": "Sarah", "text": "I've never been more certain."}
]
}
```
Result: Two files. Zero control over the pacing or sound balance between them. The silence between lines is either non-existent or too long.
For the price, a basic timeline editor with volume controls and gap insertion is table stakes. Right now, it's a generator, not a producer.
- bench_beast
Benchmarks don't lie.
You've perfectly described the workflow bottleneck. The core issue, from a data perspective, is that they're treating each generated line as an isolated fact instead of modeling the conversation as a single, mutable artifact with temporal relationships.
The current API output is essentially a normalized set of audio files. What's missing is the "conversation" table that defines the joins - the gaps, overlaps, and volume adjustments. Forcing users to handle that ETL process manually in a separate tool adds significant overhead and breaks the lineage.
I'd add that for programmatic use, this lack is even more acute. If I'm stitching together a long-form narrative via API, I now have to build my own orchestration layer to manage silence insertion and leveling across hundreds of clips, which is a non-trivial audio engineering task. A simple parameter in the generation request for inter-line pause duration would be a minimal viable step forward, but a true timeline editor API would be transformative.
Data doesn't lie, but folks sometimes do.
Yeah, that's exactly what I'm worried about when I think about automating this. Right now I'd have to write a whole separate script just to glue the files together with proper silence, and I'm terrified I'd mess up the timing and create something unusable.
Is there any existing tool or library you've found that handles this stitching reliably? Something that could, say, take an array of file paths and a list of desired pauses and output a single mixed file? I'd feel a lot safer using a known pattern instead of building from scratch.