use simple request/reply mode, instead of having multiple workers
Browse files- tts/gradio_api.py +32 -34
tts/gradio_api.py
CHANGED
@@ -33,38 +33,44 @@ def forward_gpu(file_content, wav_path, latent_file, inp_text, time_step, p_w, t
|
|
33 |
return wav_bytes
|
34 |
|
35 |
def model_worker(input_queue, output_queue, device_id):
|
36 |
-
while True:
|
37 |
-
task = input_queue.get()
|
38 |
-
inp_audio_path, inp_npy_path, inp_text, infer_timestep, p_w, t_w = task
|
39 |
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
|
|
|
|
|
|
63 |
|
64 |
|
65 |
def main(inp_audio, inp_npy, inp_text, infer_timestep, p_w, t_w, processes, input_queue, output_queue):
|
|
|
66 |
print("Push task to the inp queue |", inp_audio, inp_npy, inp_text, infer_timestep, p_w, t_w)
|
67 |
input_queue.put((inp_audio, inp_npy, inp_text, infer_timestep, p_w, t_w))
|
|
|
|
|
|
|
|
|
|
|
68 |
res = output_queue.get()
|
69 |
if res is not None:
|
70 |
return res
|
@@ -78,16 +84,8 @@ if __name__ == '__main__':
|
|
78 |
|
79 |
num_workers = 1
|
80 |
devices = [0]
|
81 |
-
input_queue = mp_manager.Queue()
|
82 |
-
output_queue = mp_manager.Queue()
|
83 |
processes = []
|
84 |
|
85 |
-
print("Start open workers")
|
86 |
-
for i in range(num_workers):
|
87 |
-
p = mp.Process(target=model_worker, args=(input_queue, output_queue, i % len(devices) if devices is not None else None))
|
88 |
-
p.start()
|
89 |
-
processes.append(p)
|
90 |
-
|
91 |
api_interface = gr.Interface(fn=
|
92 |
partial(main, processes=processes, input_queue=input_queue,
|
93 |
output_queue=output_queue),
|
|
|
33 |
return wav_bytes
|
34 |
|
35 |
def model_worker(input_queue, output_queue, device_id):
|
|
|
|
|
|
|
36 |
|
37 |
+
task = input_queue.get()
|
38 |
+
inp_audio_path, inp_npy_path, inp_text, infer_timestep, p_w, t_w = task
|
39 |
+
|
40 |
+
if inp_npy_path is None or inp_audio_path is None:
|
41 |
+
output_queue.put(None)
|
42 |
+
raise gr.Error("Please provide .wav and .npy file")
|
43 |
+
if (inp_audio_path.split('/')[-1][:-4] != inp_npy_path.split('/')[-1][:-4]):
|
44 |
+
output_queue.put(None)
|
45 |
+
raise gr.Error(".npy and .wav mismatch")
|
46 |
+
if len(inp_text) > 200:
|
47 |
+
output_queue.put(None)
|
48 |
+
raise gr.Error("input text is too long")
|
49 |
+
|
50 |
+
try:
|
51 |
+
convert_to_wav(inp_audio_path)
|
52 |
+
wav_path = os.path.splitext(inp_audio_path)[0] + '.wav'
|
53 |
+
cut_wav(wav_path, max_len=24)
|
54 |
+
with open(wav_path, 'rb') as file:
|
55 |
+
file_content = file.read()
|
56 |
+
wav_bytes = forward_gpu(file_content, wav_path, inp_npy_path, inp_text, time_step=infer_timestep, p_w=p_w, t_w=t_w)
|
57 |
+
output_queue.put(wav_bytes)
|
58 |
+
except Exception as e:
|
59 |
+
traceback.print_exc()
|
60 |
+
print(task, str(e))
|
61 |
+
output_queue.put(None)
|
62 |
+
raise gr.Error("Generation failed")
|
63 |
|
64 |
|
65 |
def main(inp_audio, inp_npy, inp_text, infer_timestep, p_w, t_w, processes, input_queue, output_queue):
|
66 |
+
input_queue = mp_manager.Queue()
|
67 |
print("Push task to the inp queue |", inp_audio, inp_npy, inp_text, infer_timestep, p_w, t_w)
|
68 |
input_queue.put((inp_audio, inp_npy, inp_text, infer_timestep, p_w, t_w))
|
69 |
+
|
70 |
+
output_queue = mp_manager.Queue()
|
71 |
+
|
72 |
+
model_worker(input_queue, output_queue, 0)
|
73 |
+
|
74 |
res = output_queue.get()
|
75 |
if res is not None:
|
76 |
return res
|
|
|
84 |
|
85 |
num_workers = 1
|
86 |
devices = [0]
|
|
|
|
|
87 |
processes = []
|
88 |
|
|
|
|
|
|
|
|
|
|
|
|
|
89 |
api_interface = gr.Interface(fn=
|
90 |
partial(main, processes=processes, input_queue=input_queue,
|
91 |
output_queue=output_queue),
|