Update pipeline/test_pipeline.py
Browse files- pipeline/test_pipeline.py +72 -0
pipeline/test_pipeline.py
CHANGED
@@ -80,3 +80,75 @@ for i in dataset:
|
|
80 |
pprint(prediction)
|
81 |
break
|
82 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
80 |
pprint(prediction)
|
81 |
break
|
82 |
|
83 |
+
print("""### P + S ###""")
|
84 |
+
pipe = pipeline(model=model_alias,
|
85 |
+
punctuator=True,
|
86 |
+
stable_ts=True,
|
87 |
+
chunk_length_s=15,
|
88 |
+
batch_size=16,
|
89 |
+
trust_remote_code=True)
|
90 |
+
dataset = load_dataset("kotoba-tech/kotoba-whisper-eval", split="train")
|
91 |
+
for i in dataset:
|
92 |
+
if i["audio"]["path"] == "long_interview_1.mp3":
|
93 |
+
i["audio"]["array"] = i["audio"]["array"][:7938000]
|
94 |
+
prediction = pipe(
|
95 |
+
i["audio"],
|
96 |
+
generate_kwargs={"language": "japanese", "task": "transcribe"}
|
97 |
+
)
|
98 |
+
pprint(prediction)
|
99 |
+
break
|
100 |
+
|
101 |
+
print("""### P ###""")
|
102 |
+
pipe = pipeline(model=model_alias,
|
103 |
+
punctuator=True,
|
104 |
+
stable_ts=False,
|
105 |
+
chunk_length_s=15,
|
106 |
+
batch_size=16,
|
107 |
+
trust_remote_code=True)
|
108 |
+
dataset = load_dataset("kotoba-tech/kotoba-whisper-eval", split="train")
|
109 |
+
for i in dataset:
|
110 |
+
if i["audio"]["path"] == "long_interview_1.mp3":
|
111 |
+
i["audio"]["array"] = i["audio"]["array"][:7938000]
|
112 |
+
prediction = pipe(
|
113 |
+
i["audio"],
|
114 |
+
generate_kwargs={"language": "japanese", "task": "transcribe"}
|
115 |
+
)
|
116 |
+
pprint(prediction)
|
117 |
+
break
|
118 |
+
|
119 |
+
print("""### S ###""")
|
120 |
+
pipe = pipeline(model=model_alias,
|
121 |
+
punctuator=False,
|
122 |
+
stable_ts=True,
|
123 |
+
chunk_length_s=15,
|
124 |
+
batch_size=16,
|
125 |
+
trust_remote_code=True)
|
126 |
+
dataset = load_dataset("kotoba-tech/kotoba-whisper-eval", split="train")
|
127 |
+
for i in dataset:
|
128 |
+
if i["audio"]["path"] == "long_interview_1.mp3":
|
129 |
+
i["audio"]["array"] = i["audio"]["array"][:7938000]
|
130 |
+
prediction = pipe(
|
131 |
+
i["audio"],
|
132 |
+
generate_kwargs={"language": "japanese", "task": "transcribe"}
|
133 |
+
)
|
134 |
+
pprint(prediction)
|
135 |
+
break
|
136 |
+
|
137 |
+
print("""### RAW ###""")
|
138 |
+
pipe = pipeline(model=model_alias,
|
139 |
+
punctuator=False,
|
140 |
+
stable_ts=False,
|
141 |
+
chunk_length_s=15,
|
142 |
+
batch_size=16,
|
143 |
+
trust_remote_code=True)
|
144 |
+
dataset = load_dataset("kotoba-tech/kotoba-whisper-eval", split="train")
|
145 |
+
for i in dataset:
|
146 |
+
if i["audio"]["path"] == "long_interview_1.mp3":
|
147 |
+
i["audio"]["array"] = i["audio"]["array"][:7938000]
|
148 |
+
prediction = pipe(
|
149 |
+
i["audio"],
|
150 |
+
generate_kwargs={"language": "japanese", "task": "transcribe"}
|
151 |
+
)
|
152 |
+
pprint(prediction)
|
153 |
+
break
|
154 |
+
|