From a65016e23e6c829d61c63a68e27b24abf86e926a Mon Sep 17 00:00:00 2001
From: zhifu gao <zhifu.gzf@alibaba-inc.com>
Date: 星期二, 16 四月 2024 10:09:03 +0800
Subject: [PATCH] Dev gzf exp (#1618)
---
funasr/models/sense_voice/model.py | 35 +++++++++++++++++++----------------
1 files changed, 19 insertions(+), 16 deletions(-)
diff --git a/funasr/models/sense_voice/model.py b/funasr/models/sense_voice/model.py
index d6552a6..4ee2fa5 100644
--- a/funasr/models/sense_voice/model.py
+++ b/funasr/models/sense_voice/model.py
@@ -73,29 +73,32 @@
speech = speech.to(device=kwargs["device"])[0, :, :]
speech_lengths = speech_lengths.to(device=kwargs["device"])
-
- task = kwargs.get("task", "ASR")
+
+ DecodingOptions = kwargs.get("DecodingOptions", {})
+ task = DecodingOptions.get("task", "ASR")
if isinstance(task, str):
task = [task]
task = "".join([f"<|{x}|>" for x in task])
initial_prompt = kwargs.get("initial_prompt", f"<|startoftranscript|>{task}")
- language = kwargs.get("language", None)
+ DecodingOptions["initial_prompt"] = initial_prompt
+
+ language = DecodingOptions.get("language", None)
language = None if language == "auto" else language
- # if language is None:
- # # detect the spoken language
- # _, probs = self.model.detect_language(speech, initial_prompt=initial_prompt)
- # print(f"Detected language: {max(probs, key=probs.get)}")
- # language = max(probs, key=probs.get)
- # language = language if kwargs.get("language", None) is None else kwargs.get("language")
-
- # decode the audio
-
- # initial_prompt = kwargs.get("initial_prompt", "<|startoftranscript|><|ASR|>")
- options = whisper.DecodingOptions(language=language, fp16=False, without_timestamps=True, initial_prompt=initial_prompt)
- result = whisper.decode(self.model, speech, options)
+ DecodingOptions["language"] = language
+ DecodingOptions["vocab_path"] = kwargs.get("vocab_path", None)
+
+
+ if "without_timestamps" not in DecodingOptions:
+ DecodingOptions["without_timestamps"] = True
+
+
+ options = whisper.DecodingOptions(**DecodingOptions)
+
+ result = whisper.decode(self.model, speech, options)
+ text = f"{result.text}"
results = []
- result_i = {"key": key[0], "text": result.text}
+ result_i = {"key": key[0], "text": text}
results.append(result_i)
--
Gitblit v1.9.1