Update handler.py
Browse files- handler.py +5 -23
handler.py
CHANGED
@@ -10,7 +10,7 @@ SAMPLE_RATE = 16000
|
|
10 |
class EndpointHandler():
|
11 |
def __init__(self, path=""):
|
12 |
self.pipeline = Pipeline.from_pretrained(
|
13 |
-
"pyannote/speaker-diarization",
|
14 |
use_auth_token=os.environ.get("HF_API_TOKEN")
|
15 |
)
|
16 |
self.pipeline.to(torch.device("cuda" if torch.cuda.is_available() else "cpu"))
|
@@ -20,12 +20,12 @@ class EndpointHandler():
|
|
20 |
Args:
|
21 |
data (Dict):
|
22 |
'inputs': Base64-encoded audio bytes
|
23 |
-
'parameters': Additional diarization parameters
|
24 |
Return:
|
25 |
Dict: Speaker diarization results
|
26 |
"""
|
27 |
inputs = data.get("inputs")
|
28 |
-
parameters = data.get("parameters", {})
|
29 |
|
30 |
# Decode the base64 audio data
|
31 |
audio_data = base64.b64decode(inputs)
|
@@ -42,27 +42,9 @@ class EndpointHandler():
|
|
42 |
|
43 |
pyannote_input = {"waveform": audio_tensor, "sample_rate": SAMPLE_RATE}
|
44 |
|
45 |
-
#
|
46 |
-
num_speakers = parameters.pop("num_speakers", None)
|
47 |
-
|
48 |
-
# Run diarization pipeline
|
49 |
try:
|
50 |
-
|
51 |
-
diarization = self.pipeline(pyannote_input, num_speakers=num_speakers) # Adjust parameters as needed for version 2.1.1
|
52 |
-
else:
|
53 |
-
diarization = self.pipeline(pyannote_input)
|
54 |
-
except TypeError as e:
|
55 |
-
print(f"Error: TypeError: {e}")
|
56 |
-
if "num_speakers" in str(e):
|
57 |
-
print("The 'num_speakers' parameter might not be supported by this version of the pipeline.")
|
58 |
-
print("Trying without num_speakers...")
|
59 |
-
try:
|
60 |
-
diarization = self.pipeline(pyannote_input)
|
61 |
-
except Exception as e:
|
62 |
-
print(f"An error occurred even without 'num_speakers': {e}")
|
63 |
-
return {"error": "Diarization failed"}
|
64 |
-
else:
|
65 |
-
return {"error": "Diarization failed with an unexpected TypeError. Check the server logs for details."}
|
66 |
except Exception as e:
|
67 |
print(f"An unexpected error occurred: {e}")
|
68 |
return {"error": "Diarization failed unexpectedly"}
|
|
|
10 |
class EndpointHandler():
|
11 |
def __init__(self, path=""):
|
12 |
self.pipeline = Pipeline.from_pretrained(
|
13 |
+
"pyannote/speaker-diarization@2.1", # 3.0 and later is nor supported as of yet in dec 2023
|
14 |
use_auth_token=os.environ.get("HF_API_TOKEN")
|
15 |
)
|
16 |
self.pipeline.to(torch.device("cuda" if torch.cuda.is_available() else "cpu"))
|
|
|
20 |
Args:
|
21 |
data (Dict):
|
22 |
'inputs': Base64-encoded audio bytes
|
23 |
+
'parameters': Additional diarization parameters (currently unused)
|
24 |
Return:
|
25 |
Dict: Speaker diarization results
|
26 |
"""
|
27 |
inputs = data.get("inputs")
|
28 |
+
parameters = data.get("parameters", {}) # We are not using them now, since model don't take speaker count anymore
|
29 |
|
30 |
# Decode the base64 audio data
|
31 |
audio_data = base64.b64decode(inputs)
|
|
|
42 |
|
43 |
pyannote_input = {"waveform": audio_tensor, "sample_rate": SAMPLE_RATE}
|
44 |
|
45 |
+
# Run diarization pipeline (without num_speakers)
|
|
|
|
|
|
|
46 |
try:
|
47 |
+
diarization = self.pipeline(pyannote_input)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
48 |
except Exception as e:
|
49 |
print(f"An unexpected error occurred: {e}")
|
50 |
return {"error": "Diarization failed unexpectedly"}
|