1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
//! Voice generation tool — `voice_gen` (`POST /v1/audio/speech`, gen-api.md §5).
//! High-quality text→speech with optional one-shot voice cloning. Synchronous;
//! the shared submit/save path lives in `tools_audio.rs`.
//!
//! Distinct from `text_to_voice` (the simple, free, offline local-OS read-aloud
//! tool) — cloning / high-quality is a separate capability, kept as its own
//! tool by design rather than overloaded onto the local TTS.
use anyhow::{Result, anyhow};
use serde_json::{Value, json};
impl super::runtime::AgentRuntime {
/// Voice generation — text → speech, with optional one-shot voice clone
/// (`reference_audio`). `text` required.
pub(crate) async fn tool_voice(&self, args: Value) -> Result<Value> {
let input = args["text"]
.as_str()
.or_else(|| args["input"].as_str())
.filter(|s| !s.is_empty())
.ok_or_else(|| anyhow!("voice_gen: `text` (the words to speak) is required"))?;
let model = args["model"]
.as_str()
.filter(|s| !s.is_empty())
.map(|m| m.rsplit('/').next().unwrap_or(m))
.unwrap_or("rsclaw-voice-v1");
let fmt = super::tools_audio::audio_format(&args);
let mut body = json!({
"model": model,
"input": input,
"response_format": fmt,
});
if let Some(voice) = args["voice"].as_str().filter(|s| !s.is_empty()) {
body["voice"] = json!(voice);
}
if let Some(instr) = args["instructions"].as_str().filter(|s| !s.is_empty()) {
body["instructions"] = json!(instr);
}
if let Some(speed) = args["speed"].as_f64() {
body["speed"] = json!(speed);
}
// One-shot voice clone: reference_audio (URL / data-URI / local path →
// base64). reference_text optionally improves fidelity.
let refs = super::tools_video::normalize_gen_assets(&args["reference_audio"]).await;
if let Some(r) = refs.first() {
body["reference_audio"] = json!({ "audio_url": r });
if let Some(rt) = args["reference_text"].as_str().filter(|s| !s.is_empty()) {
body["reference_text"] = json!(rt);
}
}
// The rsclaw voice backend REQUIRES a voice descriptor or a clone
// reference — plain text-only synthesis 502s with "provide --voice-desc
// or --ref-audio". When the caller gave neither, supply a sensible
// default voice so a bare "read this aloud" still works.
if body.get("voice").is_none() && body.get("reference_audio").is_none() {
body["voice"] = json!("标准女声，自然清晰");
}
self.audio_submit("/v1/audio/speech", &body, &fmt, "voice")
.await
}
}