batchalign 0.7.17.post1__tar.gz → 0.7.17.post3__tar.gz

{batchalign-0.7.17.post1/batchalign.egg-info → batchalign-0.7.17.post3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.7.17.post1
+Version: 0.7.17.post3
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

{batchalign-0.7.17.post1 → batchalign-0.7.17.post3}/batchalign/models/whisper/infer_asr.py

@@ -84,16 +84,28 @@ class WhisperASRModel(object):
                 [10, 5]
             ]
 
-        self.pipe = pipeline(
-            "automatic-speech-recognition",
-            model=model,
-            tokenizer=WhisperTokenizer.from_pretrained(base),
-            chunk_length_s=25,
-            stride_length_s=3,
-            device=DEVICE,
-            torch_dtype=torch.float32,
-            return_timestamps="word",
-        )
+        try:
+            self.pipe = pipeline(
+                "automatic-speech-recognition",
+                model=model,
+                tokenizer=WhisperTokenizer.from_pretrained(base),
+                chunk_length_s=25,
+                stride_length_s=3,
+                device=DEVICE,
+                torch_dtype=torch.bfloat16,
+                return_timestamps="word",
+            )
+        except TypeError:
+            self.pipe = pipeline(
+                "automatic-speech-recognition",
+                model=model,
+                tokenizer=WhisperTokenizer.from_pretrained(base),
+                chunk_length_s=25,
+                stride_length_s=3,
+                device=DEVICE,
+                torch_dtype=torch.float16,
+                return_timestamps="word",
+            )
         L.debug("Done, initalizing processor and config...")
         processor = WhisperProcessor.from_pretrained(base)
         L.debug("Whisper initialization done.")

{batchalign-0.7.17.post1 → batchalign-0.7.17.post3}/batchalign/pipelines/asr/utils.py

@@ -98,18 +98,12 @@ def retokenize_with_engine(intermediate_output, engine):
             tmp = []
 
             for s in new_ut:
-                if current_index < len(utterance):
-                    # Use current element and move index forward
-                    tmp.append((s, utterance[current_index][1]))
-                    current_index += 1
-                else:
-                    # Append with default timestamp if utterance is exhausted
-                    tmp.append((s, [None, None]))
-                    
-            if current_index >= len(utterance):
-                tmp.append((delim, [None, None]))  # Append the punctuation
+                try:
+                    tmp.append((s, utterance.pop(0)[1]))
+                except IndexError:
+                    continue
 
-            final_outputs.append((speaker, tmp))
+            final_outputs.append((speaker, tmp+[[delim, [None, None]]]))
 
     return final_outputs
 

batchalign-0.7.17.post3/batchalign/version

@@ -0,0 +1,3 @@
+0.7.17-post.3
+March 7th, 2025
+bfloat16 inference for whisper

{batchalign-0.7.17.post1 → batchalign-0.7.17.post3/batchalign.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: batchalign
-Version: 0.7.17.post1
+Version: 0.7.17.post3
 Summary: Python Speech Language Sample Analysis
 Author: Brian MacWhinney, Houjun Liu
 Author-email: macw@cmu.edu, houjun@cmu.edu

batchalign-0.7.17.post1/batchalign/version

@@ -1,3 +0,0 @@
-0.7.17-post.1
-March 6th, 2025
-Warning about ASR