[Bugfix][Frontend] Return transcription and translation verbose as float (#49073)

Signed-off-by: Lucas Christian <lucaschgf7@gmail.com>

vllm/entrypoints/speech_to_text/base/serving.py

@@ -644,7 +644,7 @@ class SpeechToTextBaseServing(GenerateBaseServing):
                         TranscriptionResponseVerbose(
                             text=text,
                             language=request.language,
-                            duration=str(duration_s),
+                            duration=duration_s,
                             segments=total_segments,
                         ),
                     )
@@ -658,7 +658,7 @@ class SpeechToTextBaseServing(GenerateBaseServing):
                         TranslationResponseVerbose(
                             text=text,
                             language=request.language,
-                            duration=str(duration_s),
+                            duration=duration_s,
                             segments=total_segments,
                         ),
                     )

vllm/entrypoints/speech_to_text/transcription/protocol.py

@@ -384,7 +384,7 @@ class TranscriptionSegment(OpenAIBaseModel):
 
 
 class TranscriptionResponseVerbose(OpenAIBaseModel):
-    duration: str
+    duration: float
     """The duration of the input audio."""
 
     language: str

vllm/entrypoints/speech_to_text/translation/protocol.py

@@ -357,7 +357,7 @@ class TranslationSegment(OpenAIBaseModel):
 
 
 class TranslationResponseVerbose(OpenAIBaseModel):
-    duration: str
+    duration: float
     """The duration of the input audio."""
 
     language: str