InputStreamにタイムスタンプ管理を追加し、STT結果出力にタイムスタンプ情報を含めるよう実装

Author: sile

src/plugin.rs

@@ -44,13 +44,15 @@ impl<'text, 'raw> TryFrom<nojson::RawJsonValue<'text, 'raw>> for SpeechToTextPlu
 struct InputStream {
     _stream_id: MediaStreamId, // TODO
     chunk: Vec<i16>,           // (left, right)...
+    timestamp: Duration,
 }
 
 impl InputStream {
     fn new(stream_id: MediaStreamId) -> Self {
         Self {
             _stream_id: stream_id,
             chunk: Vec::new(),
+            timestamp: Duration::MAX, // sentinel value
         }
     }
 
@@ -104,6 +106,8 @@ impl SpeechToTextPlugin {
     fn execute_stt(
         &mut self,
         input_stream_id: MediaStreamId,
+        timestamp: Duration,
+        duration: Duration,
         chunk: Vec<i16>,
     ) -> orfail::Result<()> {
         let request_id: usize = 0;
@@ -141,7 +145,18 @@ impl SpeechToTextPlugin {
             .or_fail()?;
         let obj = JsonObject::new(result).or_fail()?;
         let text: String = obj.get_required("text").or_fail()?;
-        todo!()
+
+        //
+        let output = nojson::object(|f| {
+            f.member("stream_id", input_stream_id.get())?;
+            f.member("timestamp", timestamp.as_secs_f32())?;
+            f.member("duration", duration.as_secs_f32())?;
+            f.member("text", &text)?;
+            Ok(())
+        });
+        writeln!(self.output_file, "{output}").or_fail()?;
+
+        Ok(())
     }
 }
 
@@ -157,18 +172,34 @@ impl MediaProcessor for SpeechToTextPlugin {
     fn process_input(&mut self, input: MediaProcessorInput) -> orfail::Result<()> {
         if let Some(sample) = input.sample {
             let input_stream = self.input_streams.get_mut(&input.stream_id).or_fail()?;
+
             let data = sample.expect_audio_data().or_fail()?;
+            if input_stream.timestamp == Duration::MAX {
+                input_stream.timestamp = data.timestamp;
+            }
+
             input_stream
                 .chunk
                 .extend(data.interleaved_stereo_samples().or_fail()?);
-            if input_stream.duration() >= self.chunk_duration {
+            let chunk_timestamp = input_stream.timestamp;
+            let chunk_duration = input_stream.duration();
+            input_stream.timestamp += chunk_duration;
+
+            if chunk_duration >= self.chunk_duration {
                 let chunk = std::mem::take(&mut input_stream.chunk);
-                self.execute_stt(input.stream_id, chunk).or_fail()?;
+                self.execute_stt(input.stream_id, chunk_timestamp, chunk_duration, chunk)
+                    .or_fail()?;
             }
         } else {
             let input_stream = self.input_streams.remove(&input.stream_id).or_fail()?;
-            self.execute_stt(input.stream_id, input_stream.chunk)
-                .or_fail()?;
+            let chunk_duration = input_stream.duration();
+            self.execute_stt(
+                input.stream_id,
+                input_stream.timestamp,
+                chunk_duration,
+                input_stream.chunk,
+            )
+            .or_fail()?;
         }
         Ok(())
     }