phostt 0.2.0

Local STT server powered by Zipformer-vi RNN-T — on-device Vietnamese speech recognition via ONNX Runtime
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
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226

//! Load tests for the phostt server.
//!
//! All tests require the Zipformer-vi ONNX model to be downloaded (~850MB).
//! Run with: `cargo test --test load_test -- --ignored`

mod common;

use futures_util::{SinkExt, StreamExt};
use std::time::{Duration, Instant};
use tokio_tungstenite::tungstenite::Message;

// ---------------------------------------------------------------------------
// 1. 4 concurrent WebSocket clients streaming 5s of audio
// ---------------------------------------------------------------------------

#[tokio::test]
#[ignore] // Requires model download — run locally
async fn test_load_4_concurrent_ws_streaming() {
    let model_dir = common::model_dir();
    let (port, _shutdown) = common::start_server(&model_dir).await;

    let start = Instant::now();
    let mut handles = Vec::new();

    for i in 0..4usize {
        handles.push(tokio::spawn(async move {
            let (mut sink, mut stream, _ready) = common::ws_connect(port).await;

            // 5s of PCM16 silence at 48kHz = 480000 bytes, sent in 9600-byte chunks
            let silence = common::generate_pcm16_silence(5.0, 48000);
            for chunk in silence.chunks(9600) {
                sink.send(Message::Binary(chunk.to_vec().into()))
                    .await
                    .unwrap_or_else(|e| panic!("Client {i}: send audio chunk failed: {e}"));
            }

            // Send Stop
            sink.send(Message::Text(
                serde_json::to_string(&serde_json::json!({"type": "stop"}))
                    .unwrap()
                    .into(),
            ))
            .await
            .unwrap_or_else(|e| panic!("Client {i}: send stop failed: {e}"));

            // Drain Partials, wait for Final
            loop {
                let msg = tokio::time::timeout(Duration::from_secs(90), stream.next())
                    .await
                    .unwrap_or_else(|_| panic!("Client {i}: timed out waiting for Final"))
                    .unwrap_or_else(|| panic!("Client {i}: stream ended before Final"))
                    .unwrap_or_else(|e| panic!("Client {i}: ws error: {e}"));

                let text = msg
                    .into_text()
                    .unwrap_or_else(|_| panic!("Client {i}: expected text message"));
                let v: serde_json::Value = serde_json::from_str(&text)
                    .unwrap_or_else(|_| panic!("Client {i}: invalid JSON: {text}"));

                match v["type"].as_str().unwrap_or("") {
                    "partial" => continue,
                    "final" => break,
                    other => panic!("Client {i}: unexpected message type: {other}"),
                }
            }

            Instant::now()
        }));
    }

    let mut error_count = 0usize;
    for (i, handle) in handles.into_iter().enumerate() {
        match tokio::time::timeout(Duration::from_secs(120), handle).await {
            Ok(Ok(_)) => {}
            Ok(Err(e)) => {
                eprintln!("Client {i} task panicked: {e}");
                error_count += 1;
            }
            Err(_) => {
                eprintln!("Client {i} timed out after 120s");
                error_count += 1;
            }
        }
    }

    let elapsed = start.elapsed();
    eprintln!("All 4 WS streaming clients completed in {elapsed:?}");
    assert_eq!(error_count, 0, "{error_count} client(s) failed");
    assert!(
        elapsed < Duration::from_secs(120),
        "Total elapsed {elapsed:?} exceeded 120s limit"
    );
}

// ---------------------------------------------------------------------------
// 2. 4 concurrent REST POST /v1/transcribe clients
// ---------------------------------------------------------------------------

#[tokio::test]
#[ignore] // Requires model download — run locally
async fn test_load_4_concurrent_rest_transcribe() {
    let model_dir = common::model_dir();
    let (port, _shutdown) = common::start_server(&model_dir).await;

    let start = Instant::now();
    let mut handles = Vec::new();

    for i in 0..4usize {
        // Generate 5s WAV at 16kHz per task
        let wav = common::generate_wav(5, 16000);

        handles.push(tokio::spawn(async move {
            let resp = reqwest::Client::new()
                .post(format!("http://127.0.0.1:{port}/v1/transcribe"))
                .body(wav)
                .send()
                .await
                .unwrap_or_else(|e| panic!("Client {i}: POST /v1/transcribe failed: {e}"));

            let status = resp.status();
            assert_eq!(status, 200, "Client {i}: expected 200, got {status}");

            i
        }));
    }

    let mut error_count = 0usize;
    for (i, handle) in handles.into_iter().enumerate() {
        match tokio::time::timeout(Duration::from_secs(120), handle).await {
            Ok(Ok(_)) => {}
            Ok(Err(e)) => {
                eprintln!("Client {i} task panicked: {e}");
                error_count += 1;
            }
            Err(_) => {
                eprintln!("Client {i} timed out after 120s");
                error_count += 1;
            }
        }
    }

    let elapsed = start.elapsed();
    eprintln!("All 4 REST transcribe clients completed in {elapsed:?}");
    assert_eq!(error_count, 0, "{error_count} client(s) failed");
    assert!(
        elapsed < Duration::from_secs(120),
        "Total elapsed {elapsed:?} exceeded 120s limit"
    );
}

// ---------------------------------------------------------------------------
// 3. Burst of 20 WebSocket connections (pool is 4, rest will queue)
// ---------------------------------------------------------------------------

#[tokio::test]
#[ignore] // Requires model download — run locally
async fn test_load_burst_20_connections() {
    let model_dir = common::model_dir();
    let (port, _shutdown) = common::start_server(&model_dir).await;

    let start = Instant::now();
    let mut handles = Vec::new();

    // Spawn all 20 tasks immediately with no delay — connections will queue
    // because the server pool is 4. That is expected behaviour.
    for i in 0..20usize {
        handles.push(tokio::spawn(async move {
            let (mut sink, mut stream, _ready) = common::ws_connect(port).await;

            // Send Stop immediately — no audio
            sink.send(Message::Text(
                serde_json::to_string(&serde_json::json!({"type": "stop"}))
                    .unwrap()
                    .into(),
            ))
            .await
            .unwrap_or_else(|e| panic!("Client {i}: send stop failed: {e}"));

            // Wait for Final
            loop {
                let msg = tokio::time::timeout(Duration::from_secs(25), stream.next())
                    .await
                    .unwrap_or_else(|_| panic!("Client {i}: timed out waiting for Final"))
                    .unwrap_or_else(|| panic!("Client {i}: stream ended before Final"))
                    .unwrap_or_else(|e| panic!("Client {i}: ws error: {e}"));

                let text = msg
                    .into_text()
                    .unwrap_or_else(|_| panic!("Client {i}: expected text message"));
                let v: serde_json::Value = serde_json::from_str(&text)
                    .unwrap_or_else(|_| panic!("Client {i}: invalid JSON: {text}"));

                match v["type"].as_str().unwrap_or("") {
                    "partial" => continue,
                    "final" => break,
                    other => panic!("Client {i}: unexpected message type: {other}"),
                }
            }

            i
        }));
    }

    let mut error_count = 0usize;
    for (i, handle) in handles.into_iter().enumerate() {
        match tokio::time::timeout(Duration::from_secs(30), handle).await {
            Ok(Ok(_)) => {}
            Ok(Err(e)) => {
                eprintln!("Client {i} task panicked: {e}");
                error_count += 1;
            }
            Err(_) => {
                eprintln!("Client {i} timed out after 30s");
                error_count += 1;
            }
        }
    }

    let elapsed = start.elapsed();
    eprintln!("All 20 burst WS connections completed in {elapsed:?}");
    assert_eq!(error_count, 0, "{error_count} client(s) failed or panicked");
    assert!(
        elapsed < Duration::from_secs(30),
        "Total elapsed {elapsed:?} exceeded 30s limit"
    );
}