Merge branch 'guillaumekln:master' into return-best-result

README.md

@@ -165,12 +165,12 @@ Models can also be converted from the code. See the [conversion API](https://ope
 
 1. Directly load the model from a local directory:
 ```python
-model = faster_whisper.WhisperModel('whisper-large-v2-ct2')
+model = faster_whisper.WhisperModel("whisper-large-v2-ct2")
 ```
 
 2. [Upload your model to the Hugging Face Hub](https://huggingface.co/docs/transformers/model_sharing#upload-with-the-web-interface) and load it from its name:
 ```python
-model = faster_whisper.WhisperModel('username/whisper-large-v2-ct2')
+model = faster_whisper.WhisperModel("username/whisper-large-v2-ct2")
 ```
 
 ## Comparing performance against other implementations

faster_whisper/transcribe.py

@@ -699,6 +699,8 @@ def get_prompt(
             prefix_tokens = tokenizer.encode(" " + prefix.strip())
             if len(prefix_tokens) >= self.max_length // 2:
                 prefix_tokens = prefix_tokens[: self.max_length // 2 - 1]
+            if not without_timestamps:
+                prompt.append(tokenizer.timestamp_begin)
             prompt.extend(prefix_tokens)
 
         return prompt
@@ -733,8 +735,6 @@ def add_word_timestamps(
         # hack: truncate long words at sentence boundaries.
         # a better segmentation algorithm based on VAD should be able to replace this.
         if len(word_durations) > 0:
-            median_duration = np.median(word_durations)
-            max_duration = median_duration * 2
             sentence_end_marks = ".。!！?？"
             # ensure words at sentence boundaries
             # are not longer than twice the median word duration.

faster_whisper/version.py

@@ -1,3 +1,3 @@
 """Version information."""
 
-__version__ = "0.6.0"
+__version__ = "0.7.0"

requirements.txt

@@ -1,5 +1,5 @@
 av==10.*
-ctranslate2>=3.10,<4
+ctranslate2>=3.17,<4
 huggingface_hub>=0.13
 tokenizers==0.13.*
 onnxruntime>=1.14,<2

tests/test_transcribe.py

@@ -34,6 +34,24 @@ def test_transcribe(jfk_path):
     assert segment.end == segment.words[-1].end
 
 
+def test_prefix_with_timestamps(jfk_path):
+    model = WhisperModel("tiny")
+    segments, _ = model.transcribe(jfk_path, prefix="And so my fellow Americans")
+    segments = list(segments)
+
+    assert len(segments) == 1
+
+    segment = segments[0]
+
+    assert segment.text == (
+        " And so my fellow Americans ask not what your country can do for you, "
+        "ask what you can do for your country."
+    )
+
+    assert segment.start == 0
+    assert 10 < segment.end < 11
+
+
 def test_vad(jfk_path):
     model = WhisperModel("tiny")
     segments, info = model.transcribe(