hoosh 1.3.0

AI inference gateway — multi-provider LLM routing, local model serving, speech-to-text, and token budget management
Documentation 
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
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136

//! Audio route handlers — speech-to-text (Whisper) and text-to-speech (Piper).

#[cfg(any(feature = "whisper", feature = "piper"))]
use std::sync::Arc;

#[cfg(any(feature = "whisper", feature = "piper"))]
use axum::{Json, extract::State, http::StatusCode, response::IntoResponse};

#[cfg(any(feature = "whisper", feature = "piper"))]
use super::AppState;
#[cfg(any(feature = "whisper", feature = "piper"))]
use super::types::error_response;

// ---------------------------------------------------------------------------
// Speech-to-text: /v1/audio/transcriptions (OpenAI-compatible)
// ---------------------------------------------------------------------------

#[cfg(feature = "whisper")]
pub(crate) async fn transcribe(
    State(state): State<Arc<AppState>>,
    headers: axum::http::HeaderMap,
    body: axum::body::Bytes,
) -> impl IntoResponse {
    if let Some(ct) = headers.get("content-type") {
        let ct_str = ct.to_str().unwrap_or("");
        if !ct_str.starts_with("audio/")
            && !ct_str.starts_with("application/octet-stream")
            && !ct_str.starts_with("multipart/form-data")
        {
            return error_response(
                StatusCode::UNSUPPORTED_MEDIA_TYPE,
                format!("expected audio/* content type, got: {ct_str}"),
            )
            .into_response();
        }
    }

    let whisper = match &state.whisper {
        Some(w) => w.clone(),
        None => {
            return error_response(
                StatusCode::SERVICE_UNAVAILABLE,
                "Whisper model not loaded. Set whisper_model in config.",
            )
            .into_response();
        }
    };

    let request = crate::inference::TranscriptionRequest {
        audio: body.to_vec(),
        language: None,
        word_timestamps: false,
    };

    match whisper.transcribe_async(request).await {
        Ok(result) => {
            let resp = serde_json::json!({
                "text": result.text,
                "language": result.language,
                "duration": result.duration_secs,
                "segments": result.segments.iter().map(|s| {
                    serde_json::json!({
                        "text": s.text,
                        "start": s.start_secs,
                        "end": s.end_secs,
                    })
                }).collect::<Vec<_>>(),
            });
            (StatusCode::OK, Json(resp)).into_response()
        }
        Err(e) => error_response(
            StatusCode::INTERNAL_SERVER_ERROR,
            format!("Transcription error: {e}"),
        )
        .into_response(),
    }
}

// ---------------------------------------------------------------------------
// Text-to-speech: /v1/audio/speech (OpenAI-compatible)
// ---------------------------------------------------------------------------

#[cfg(feature = "piper")]
pub(crate) async fn text_to_speech(
    State(state): State<Arc<AppState>>,
    Json(req): Json<crate::inference::SpeechRequest>,
) -> impl IntoResponse {
    let tts = match &state.tts {
        Some(t) => t.clone(),
        None => {
            return error_response(
                StatusCode::SERVICE_UNAVAILABLE,
                "TTS model not loaded. Set tts_model in config.",
            )
            .into_response();
        }
    };

    if req.input.is_empty() {
        return error_response(StatusCode::BAD_REQUEST, "input text is required").into_response();
    }
    if req.input.len() > 4096 {
        return error_response(
            StatusCode::BAD_REQUEST,
            format!("input too long: {} chars (max 4096)", req.input.len()),
        )
        .into_response();
    }
    if !(0.25..=4.0).contains(&req.speed) {
        return error_response(
            StatusCode::BAD_REQUEST,
            format!("speed must be between 0.25 and 4.0, got {}", req.speed),
        )
        .into_response();
    }

    match tts.synthesize(&req).await {
        Ok(result) => {
            let content_type = match result.format.as_str() {
                "pcm" => "audio/pcm",
                _ => "audio/wav",
            };
            (
                StatusCode::OK,
                [(axum::http::header::CONTENT_TYPE, content_type)],
                result.audio,
            )
                .into_response()
        }
        Err(e) => error_response(
            StatusCode::INTERNAL_SERVER_ERROR,
            format!("TTS synthesis error: {e}"),
        )
        .into_response(),
    }
}