Adding timestamps for CTC with LM in ASR pipeline. (#15863)

* Adding timestamps for CTC with LM in ASR pipeline.

* iRemove print.

* Nit change.

src/transformers/models/wav2vec2/tokenization_wav2vec2.py

@@ -353,7 +353,7 @@ class Wav2Vec2CTCTokenizer(PreTrainedTokenizer):
                     word = char
 
             last_state = state
-        if state == "WORD":
+        if last_state == "WORD":
             word_offsets.append({"word": word, "start_offset": start_offset, "end_offset": end_offset})
 
         return word_offsets

src/transformers/pipelines/automatic_speech_recognition.py

@@ -313,8 +313,10 @@ class AutomaticSpeechRecognitionPipeline(ChunkPipeline):
         # Optional return types
         optional = {}
 
-        if return_timestamps and self.type != "ctc":
+        if return_timestamps and self.type == "seq2seq":
             raise ValueError("We cannot return_timestamps yet on non-ctc models !")
+        if return_timestamps == "char" and self.type == "ctc_with_lm":
+            raise ValueError("CTC with LM cannot return `char` timestamps, only `words`")
 
         final_items = []
         key = "logits" if self.type == "ctc_with_lm" else "tokens"
@@ -335,32 +337,41 @@ class AutomaticSpeechRecognitionPipeline(ChunkPipeline):
         if self.type == "ctc_with_lm":
             if decoder_kwargs is None:
                 decoder_kwargs = {}
-            text = self.decoder.decode_beams(items, **decoder_kwargs)[0][0]
+            beams = self.decoder.decode_beams(items, **decoder_kwargs)
+            text = beams[0][0]
+            if return_timestamps:
+                # Simply cast from pyctcdecode format to wav2vec2 format to leverage
+                # pre-existing code later
+                chunk_offset = beams[0][2]
+                word_offsets = []
+                for word, (start_offset, end_offset) in chunk_offset:
+                    word_offsets.append({"word": word, "start_offset": start_offset, "end_offset": end_offset})
 
         else:
             skip_special_tokens = self.type != "ctc"
             text = self.tokenizer.decode(items, skip_special_tokens=skip_special_tokens)
             if return_timestamps:
-                if return_timestamps == "char":
-                    decoded = self.tokenizer.decode(
+                char_offsets = self.tokenizer.decode(
                     items, skip_special_tokens=skip_special_tokens, output_char_offsets=True
+                )["char_offsets"]
+                if return_timestamps == "word":
+                    word_offsets = self.tokenizer._get_word_offsets(
+                        char_offsets, self.tokenizer.replace_word_delimiter_char
                     )
-                elif return_timestamps == "word":
-                    decoded = self.tokenizer.decode(
-                        items, skip_special_tokens=skip_special_tokens, output_word_offsets=True
-                    )
+
+        if return_timestamps:
+            if return_timestamps == "word":
+                offsets = word_offsets
+            else:
+                offsets = char_offsets
             chunks = []
-                for item in decoded[f"{return_timestamps}_offsets"]:
-                    start = (
-                        item["start_offset"]
-                        * self.model.config.inputs_to_logits_ratio
-                        / self.feature_extractor.sampling_rate
-                    )
-                    stop = (
-                        item["end_offset"]
-                        * self.model.config.inputs_to_logits_ratio
-                        / self.feature_extractor.sampling_rate
-                    )
+            for item in offsets:
+                start = item["start_offset"] * self.model.config.inputs_to_logits_ratio
+                start /= self.feature_extractor.sampling_rate
+
+                stop = item["end_offset"] * self.model.config.inputs_to_logits_ratio
+                stop /= self.feature_extractor.sampling_rate
+
                 chunks.append({"text": item[return_timestamps], "timestamp": (start, stop)})
             optional["chunks"] = chunks
 

tests/pipelines/test_pipelines_automatic_speech_recognition.py

@@ -188,6 +188,32 @@ class AutomaticSpeechRecognitionPipelineTests(unittest.TestCase, metaclass=Pipel
             },
         )
 
+        speech_recognizer.type = "ctc_with_lm"
+        # Simple test with CTC with LM, chunking + timestamps
+        output = speech_recognizer(filename, chunk_length_s=2.0, return_timestamps="word")
+        self.assertEqual(
+            output,
+            {
+                "text": "y en las ramas medio sumergidas revoloteaban algunos pájaros de quimérico y legendario plumajcri",
+                "chunks": [
+                    {"text": "y", "timestamp": (0.52, 0.54)},
+                    {"text": "en", "timestamp": (0.6, 0.68)},
+                    {"text": "las", "timestamp": (0.74, 0.84)},
+                    {"text": "ramas", "timestamp": (0.94, 1.24)},
+                    {"text": "medio", "timestamp": (1.32, 1.52)},
+                    {"text": "sumergidas", "timestamp": (1.56, 2.22)},
+                    {"text": "revoloteaban", "timestamp": (2.36, 3.0)},
+                    {"text": "algunos", "timestamp": (3.06, 3.38)},
+                    {"text": "pájaros", "timestamp": (3.46, 3.86)},
+                    {"text": "de", "timestamp": (3.92, 4.0)},
+                    {"text": "quimérico", "timestamp": (4.08, 4.6)},
+                    {"text": "y", "timestamp": (4.66, 4.68)},
+                    {"text": "legendario", "timestamp": (4.74, 5.26)},
+                    {"text": "plumajcri", "timestamp": (5.34, 5.74)},
+                ],
+            },
+        )
+
     @require_tf
     def test_small_model_tf(self):
         self.skipTest("Tensorflow not supported yet.")