Update handler.py
Browse files- handler.py +12 -10
handler.py
CHANGED
@@ -8,20 +8,21 @@ SAMPLE_RATE = 16000
|
|
8 |
MODEL_NAME = "openai/whisper-large" #this always needs to stay in line 8 :D sorry for the hackiness
|
9 |
lang = "dk"
|
10 |
|
11 |
-
device = 0 if torch.cuda.is_available() else "cpu"
|
12 |
-
pipe = pipeline(
|
13 |
-
task="automatic-speech-recognition",
|
14 |
-
model=MODEL_NAME,
|
15 |
-
chunk_length_s=30,
|
16 |
-
device=device,
|
17 |
-
)
|
18 |
|
19 |
-
|
20 |
|
21 |
class EndpointHandler():
|
22 |
def __init__(self, path=""):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
# load the model
|
24 |
-
self.model = whisper.load_model("
|
|
|
25 |
|
26 |
|
27 |
def __call__(self, data: Dict[str, bytes]) -> Dict[str, str]:
|
@@ -37,8 +38,9 @@ class EndpointHandler():
|
|
37 |
audio_nparray = ffmpeg_read(inputs, SAMPLE_RATE)
|
38 |
audio_tensor= torch.from_numpy(audio_nparray)
|
39 |
|
40 |
-
# run inference pipeline
|
41 |
result = self.model.transcribe(audio_nparray)
|
42 |
|
|
|
43 |
# postprocess the prediction
|
44 |
return {"tekst": result["text"]}
|
|
|
8 |
MODEL_NAME = "openai/whisper-large" #this always needs to stay in line 8 :D sorry for the hackiness
|
9 |
lang = "dk"
|
10 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
11 |
|
12 |
+
|
13 |
|
14 |
class EndpointHandler():
|
15 |
def __init__(self, path=""):
|
16 |
+
pipe = pipeline(
|
17 |
+
task="automatic-speech-recognition",
|
18 |
+
model=MODEL_NAME,
|
19 |
+
chunk_length_s=30,
|
20 |
+
device=device,
|
21 |
+
)
|
22 |
+
|
23 |
# load the model
|
24 |
+
#self.model = whisper.load_model("large")
|
25 |
+
self.model.config.forced_decoder_ids = pipe.tokenizer.get_decoder_prompt_ids(language=lang, task="transcribe")
|
26 |
|
27 |
|
28 |
def __call__(self, data: Dict[str, bytes]) -> Dict[str, str]:
|
|
|
38 |
audio_nparray = ffmpeg_read(inputs, SAMPLE_RATE)
|
39 |
audio_tensor= torch.from_numpy(audio_nparray)
|
40 |
|
41 |
+
# run inference pipeline
|
42 |
result = self.model.transcribe(audio_nparray)
|
43 |
|
44 |
+
|
45 |
# postprocess the prediction
|
46 |
return {"tekst": result["text"]}
|