inference/tts: return InferenceResult struct

Signed-off-by: Stijn Tintel <stijn@linux-ipv6.be>

Author: stintel

src/inference/tts.rs

@@ -3,6 +3,9 @@ use std::sync::{Arc, Mutex};
 
 use anyhow::{Result, anyhow};
 use sherpa_rs::tts::{TtsAudio, VitsTts, VitsTtsConfig};
+use tokio::time::Instant;
+
+use crate::inference::InferenceResult;
 
 #[derive(Clone)]
 pub struct TtsEngine {
@@ -34,12 +37,39 @@ impl TtsEngine {
 
     /// # Errors
     /// - when the mutex is poisoned
-    pub fn synthesize(&self, text: &str, sid: i32, speed: f32) -> Result<TtsAudio> {
+    pub fn synthesize(
+        &self,
+        text: &str,
+        sid: i32,
+        speed: f32,
+    ) -> Result<InferenceResult<TtsAudio>> {
         let mut tts = self
             .tts
             .lock()
             .map_err(|e| anyhow!("TTS mutex poisoned: {e:#?}"))?;
 
-        tts.create(text, sid, speed).map_err(|e| anyhow!("{e:#?}"))
+        let start = Instant::now();
+
+        let speech = tts
+            .create(text, sid, speed)
+            .map_err(|e| anyhow!("{e:#?}"))?;
+
+        let time = start.elapsed().as_secs_f64();
+        let time_ms = time * 1000.0;
+        #[allow(clippy::cast_precision_loss)]
+        let speedup = if time_ms > 0.0 {
+            (f64::from(speech.duration)) / time_ms
+        } else {
+            0.0
+        };
+
+        let result = InferenceResult {
+            duration: u64::try_from(speech.duration).unwrap_or(0),
+            output: speech,
+            speedup,
+            time,
+        };
+
+        Ok(result)
     }
 }

src/routes/api/tts.rs

@@ -38,6 +38,7 @@ pub async fn get(
             .into_response();
     };
 
+    let text = parameters.text.clone();
     let tts_audio = spawn_blocking(move || tts_engine.synthesize(&parameters.text, 0, 1.0)).await;
 
     let Ok(Ok(tts_audio)) = tts_audio else {
@@ -48,7 +49,15 @@ pub async fn get(
             .into_response();
     };
 
-    let Ok(audio_bytes) = encode_wav(&tts_audio.samples, tts_audio.sample_rate) else {
+    tracing::info!(
+        "inference took {}s: {} - speedup: {}x",
+        tts_audio.time,
+        text,
+        tts_audio.speedup
+    );
+
+    let Ok(audio_bytes) = encode_wav(&tts_audio.output.samples, tts_audio.output.sample_rate)
+    else {
         return (
             StatusCode::INTERNAL_SERVER_ERROR,
             String::from("failed to encode audio in WAV"),