Update app.py

app.py

@@ -3,20 +3,27 @@ import gradio as gr
 from transformers import pipeline
 
 title = "Transcribe speech several languages"
+device = "cuda:0" if torch.cuda.is_available() else "cpu"
 
-pipelineGE = pipeline(task="automatic-speech-recognition", model="jonatasgrosman/wav2vec2-large-xlsr-53-german")
-pipelineEN = pipeline(task="automatic-speech-recognition", model="openai/whisper-large")
+asr_pipe_audio2Text_Ge = pipeline(task="automatic-speech-recognition", model="jonatasgrosman/wav2vec2-large-xlsr-53-german")
+asr_pipe_whisper = pipeline(task="automatic-speech-recognition", model="openai/whisper-large", device=device)
 
-def transcribeFile(audio_path : str) -> str:
-    transcription = pipelineGE(audio_path)
-    return transcription["text"]
+#def transcribeFile(audio_path : str) -> str:
+#    transcription = asr_pipe_audio2Text_Ge(audio_path)
+#   return transcription["text"]
+
+def translateAudio(audio_path):
+    translationOutput = asr_pipe_whisper(audio_path, max_new_tokens=256, generate_kwargs={"task":"translate"})
+    return translationOutput["text"]
 
 def transcribeFileMulti(inputlang, audio_path : str) -> str:
     if inputlang == "English":
-        transcription = pipelineEN(audio_path)
+        transcription = asr_pipe_whisper(audio_path)
     elif inputlang == "German":
-        transcription = pipelineGE(audio_path)
-    return transcription["text"]
+        transcription = asr_pipe_audio2Text_Ge(audio_path)
+        translation = translateAudio(audio_path)
+        output = transcription + translation
+    return output #transcription["text"]