aurum_core/remote/
openai_speech.rs

1//! OpenAI-compatible audio speech request protocol (JOE-1939).
2//!
3//! Encodes documented speech request fields without credentials or endpoint
4//! logic. Shared by OpenRouter TTS and future OpenAI first-party TTS.
5
6use serde::{Deserialize, Serialize};
7
8/// Wire `response_format` values for `/audio/speech`.
9#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
10#[serde(rename_all = "snake_case")]
11pub enum SpeechResponseFormat {
12    /// Raw s16le PCM (preferred for Aurum normalization).
13    Pcm,
14    /// MPEG Layer III (normalized via supervised FFmpeg).
15    Mp3,
16}
17
18impl SpeechResponseFormat {
19    pub fn as_str(self) -> &'static str {
20        match self {
21            Self::Pcm => "pcm",
22            Self::Mp3 => "mp3",
23        }
24    }
25}
26
27/// OpenAI-compatible speech request body (no secrets).
28#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
29pub struct OpenAiSpeechRequest {
30    pub model: String,
31    pub input: String,
32    pub voice: String,
33    #[serde(skip_serializing_if = "Option::is_none")]
34    pub response_format: Option<SpeechResponseFormat>,
35    #[serde(skip_serializing_if = "Option::is_none")]
36    pub speed: Option<f64>,
37}
38
39impl OpenAiSpeechRequest {
40    /// Build a request, omitting unsupported/default fields when helpful.
41    pub fn new(
42        model: impl Into<String>,
43        input: impl Into<String>,
44        voice: impl Into<String>,
45        format: SpeechResponseFormat,
46        speed: Option<f32>,
47    ) -> Self {
48        Self {
49            model: model.into(),
50            input: input.into(),
51            voice: voice.into(),
52            response_format: Some(format),
53            speed: speed.map(|s| s as f64),
54        }
55    }
56
57    pub fn to_json_bytes(&self) -> Result<Vec<u8>, serde_json::Error> {
58        serde_json::to_vec(self)
59    }
60}
61
62/// Parse sample rate from `Content-Type` like `audio/pcm;rate=24000;channels=1`.
63///
64/// Returns `(sample_rate_hz, channels)` when both present; defaults channels to 1
65/// when only rate is present.
66pub fn parse_pcm_content_type(content_type: &str) -> Option<(u32, u16)> {
67    let ct = content_type.to_ascii_lowercase();
68    if !ct.contains("audio/pcm") && !ct.starts_with("audio/l16") {
69        return None;
70    }
71    let mut rate = None;
72    let mut channels = None;
73    for part in ct.split(';').skip(1) {
74        let part = part.trim();
75        if let Some(v) = part.strip_prefix("rate=") {
76            rate = v.trim().parse().ok();
77        } else if let Some(v) = part.strip_prefix("channels=") {
78            channels = v.trim().parse().ok();
79        }
80    }
81    let rate = rate?;
82    Some((rate, channels.unwrap_or(1)))
83}
84
85#[cfg(test)]
86mod tests {
87    use super::*;
88
89    #[test]
90    fn serializes_speech_request() {
91        let req = OpenAiSpeechRequest::new(
92            "openai/gpt-4o-mini-tts",
93            "Hello",
94            "alloy",
95            SpeechResponseFormat::Pcm,
96            Some(1.0),
97        );
98        let v: serde_json::Value = serde_json::from_slice(&req.to_json_bytes().unwrap()).unwrap();
99        assert_eq!(v["model"], "openai/gpt-4o-mini-tts");
100        assert_eq!(v["input"], "Hello");
101        assert_eq!(v["voice"], "alloy");
102        assert_eq!(v["response_format"], "pcm");
103        assert_eq!(v["speed"], 1.0);
104        // No secrets ever.
105        let s = v.to_string();
106        assert!(!s.contains("Authorization"));
107        assert!(!s.contains("api_key"));
108    }
109
110    #[test]
111    fn omits_speed_when_none() {
112        let req = OpenAiSpeechRequest::new("m", "i", "v", SpeechResponseFormat::Mp3, None);
113        let s = serde_json::to_string(&req).unwrap();
114        assert!(!s.contains("speed"));
115        assert!(s.contains("mp3"));
116    }
117
118    #[test]
119    fn parse_pcm_content_type_rate_channels() {
120        assert_eq!(
121            parse_pcm_content_type("audio/pcm;rate=24000;channels=1"),
122            Some((24_000, 1))
123        );
124        assert_eq!(
125            parse_pcm_content_type("AUDIO/PCM; rate=16000"),
126            Some((16_000, 1))
127        );
128        assert!(parse_pcm_content_type("audio/mpeg").is_none());
129    }
130}
aurum_core/remote/openai_speech.rs

aurum_core/remote/
openai_speech.rs
