Better logic to update the voices form assigned_voice and speaker_id

open_dubbing/dubbing.py

@@ -393,6 +393,7 @@ def update(self):
         )
 
         modified_utterances = self.tts.update_utterance_metadata(
+            utterance=utterance,
             utterance_metadata=modified_utterances,
             assigned_voices=assigned_voices,
         )

open_dubbing/text_to_speech.py

@@ -21,6 +21,7 @@
 from pydub import AudioSegment
 
 from open_dubbing.ffmpeg import FFmpeg
+from open_dubbing.utterance import Utterance
 
 
 class Voice(NamedTuple):
@@ -106,18 +107,30 @@ def _add_text_to_speech_properties(
     def update_utterance_metadata(
         self,
         *,
+        utterance: Utterance | None = None,
         utterance_metadata: Sequence[Mapping[str, str | float]],
         assigned_voices: Mapping[str, str] | None,
     ) -> Sequence[Mapping[str, str | float]]:
         """Updates utterance metadata with assigned voices."""
         updated_utterance_metadata = []
         for metadata_item in utterance_metadata:
             new_utterance = metadata_item.copy()
-            speaker_id = new_utterance.get("speaker_id")
-            new_utterance["assigned_voice"] = assigned_voices.get(speaker_id)
-            new_utterance = self._add_text_to_speech_properties(
-                utterance_metadata=new_utterance
+
+            fields = (
+                utterance.get_modified_utterance_fields(new_utterance)
+                if utterance
+                else []
             )
+            # If "assigned_voice" has changed we give it priority and not overwrite it
+            # by recalculating from speaker_id/gender
+            if not utterance or (
+                "speaker_id" in fields and "assigned_voice" not in fields
+            ):
+                speaker_id = new_utterance.get("speaker_id")
+                new_utterance["assigned_voice"] = assigned_voices.get(speaker_id)
+                new_utterance = self._add_text_to_speech_properties(
+                    utterance_metadata=new_utterance
+                )
             updated_utterance_metadata.append(new_utterance)
         return updated_utterance_metadata
 

open_dubbing/utterance.py

@@ -102,11 +102,25 @@ def save_utterances(
         except Exception as e:
             logging.warning(f"Error saving utterance metadata: {e}")
 
+    def _get_utterance_fields_to_hash(self, utterance):
+        filtered_fields = {
+            key: value for key, value in utterance.items() if not key.startswith("_")
+        }
+        return filtered_fields
+
     def _hash_utterances(self, utterance_metadata):
         for utterance in utterance_metadata:
-            dict_str = json.dumps(utterance, sort_keys=True)
+            filtered_fields = self._get_utterance_fields_to_hash(utterance)
+            dict_str = json.dumps(filtered_fields, sort_keys=True)
             _hash = hashlib.sha256(dict_str.encode()).hexdigest()
-            utterance["hash"] = _hash
+            utterance["_hash"] = _hash
+
+            for field in ["assigned_voice", "speaker_id"]:
+                value = utterance.get(field)
+                if value:
+                    utterance[f"_{field}_hash"] = hashlib.sha256(
+                        value.encode()
+                    ).hexdigest()
 
         return utterance_metadata
 
@@ -129,13 +143,28 @@ def get_files_paths(self, utterance_metadata) -> Tuple[List[str], List[str]]:
 
         return paths, dubbed_paths
 
+    def get_modified_utterance_fields(self, utterance):
+        modified = []
+        for field in utterance:
+            field_hash = utterance.get(f"_{field}_hash")
+            if not field_hash:
+                continue
+
+            field_value = utterance[field]
+            current_hash = hashlib.sha256(field_value.encode()).hexdigest()
+
+            if current_hash != field_hash:
+                modified.append(field)
+
+        return modified
+
     def get_modified_utterances(self, utterance_metadata):
         modified = []
         for utterance in utterance_metadata:
-            _hash_utterance = utterance["hash"]
-            del utterance["hash"]
-            dict_str = json.dumps(utterance, sort_keys=True)
+            _hash_utterance = utterance["_hash"]
+            filtered_fields = self._get_utterance_fields_to_hash(utterance)
 
+            dict_str = json.dumps(filtered_fields, sort_keys=True)
             _hash = hashlib.sha256(dict_str.encode()).hexdigest()
             if _hash_utterance != _hash:
                 modified.append(utterance)

tests/text_to_speech_test.py

@@ -17,7 +17,7 @@
 import tempfile
 
 from typing import List
-from unittest.mock import patch
+from unittest.mock import Mock, patch
 
 import pytest
 
@@ -126,7 +126,7 @@ def test_dub_utterances_with_speeds(
                 "start": 0,
                 "end": 5,
                 "translated_text": "Hello world",
-                "speed": 1.0,  # Initially set speed to 1.0
+                "speed": 1.0,
                 "path": "some/path/file.mp3",
             }
         ]
@@ -292,3 +292,61 @@ def test_assign_voices(self):
                 target_language_region="IN",
             )
             assert {1: "Voice3"} == results
+
+    def _get_update_utterance_metadata(self):
+        return [
+            {
+                "speaker_id": "2",
+                "start": 0,
+                "assigned_voice": "Voice0",
+                "end": 5,
+                "translated_text": "Hello world",
+                "speed": 1.0,
+                "path": "some/path/file.mp3",
+            }
+        ]
+
+    def test_update_utterance_metadata_assign_voice_from_speaker(self):
+        voices = {"1": "Voice1", "2": "Voice2"}
+
+        utterance_metadata = self._get_update_utterance_metadata()
+        tts = TextToSpeechUT()
+        updated_utterances = tts.update_utterance_metadata(
+            utterance=None,
+            utterance_metadata=utterance_metadata,
+            assigned_voices=voices,
+        )
+
+        assert "Voice2" == updated_utterances[0]["assigned_voice"]
+
+    def test_update_utterance_metadata_assign_voice_from_speaker_id(self):
+        voices = {"1": "Voice1", "2": "Voice2"}
+
+        utterance_metadata = self._get_update_utterance_metadata()
+        utterance = Mock()
+        utterance.get_modified_utterance_fields.return_value = {"speaker_id"}
+
+        tts = TextToSpeechUT()
+        updated_utterances = tts.update_utterance_metadata(
+            utterance=utterance,
+            utterance_metadata=utterance_metadata,
+            assigned_voices=voices,
+        )
+
+        assert "Voice2" == updated_utterances[0]["assigned_voice"]
+
+    def test_update_utterance_metadata_assign_voice_from_assigned_voice(self):
+        voices = {"1": "Voice1", "2": "Voice2"}
+
+        utterance_metadata = self._get_update_utterance_metadata()
+        utterance = Mock()
+        utterance.get_modified_utterance_fields.return_value = {"assigned_voice"}
+
+        tts = TextToSpeechUT()
+        updated_utterances = tts.update_utterance_metadata(
+            utterance=utterance,
+            utterance_metadata=utterance_metadata,
+            assigned_voices=voices,
+        )
+
+        assert "Voice0" == updated_utterances[0]["assigned_voice"]

tests/utterance_test.py

@@ -50,13 +50,13 @@ def testrun_save_utterance(self):
                             "id": 1,
                             "start": 1.26,
                             "end": 3.94,
-                            "hash": "26d514d9ce21021f51bd010d9946db0f31555ef7145067d4fe5a3b1bdcd84ce7",
+                            "_hash": "26d514d9ce21021f51bd010d9946db0f31555ef7145067d4fe5a3b1bdcd84ce7",
                         },
                         {
                             "id": 2,
                             "start": 5.24,
                             "end": 6.629,
-                            "hash": "157dc7fb355c7dc13a0ea687e9fd4a6f6c5c03526a959a64dfe1fa7562fedff4",
+                            "_hash": "157dc7fb355c7dc13a0ea687e9fd4a6f6c5c03526a959a64dfe1fa7562fedff4",
                         },
                     ],
                     "metadata": {
@@ -70,6 +70,7 @@ def test_hash_utterances(self):
             {
                 "start": 1.26,
                 "end": 3.94,
+                "_private_not_hashed": "0",
             },
             {
                 "start": 5.24,
@@ -86,12 +87,13 @@ def test_hash_utterances(self):
             {
                 "start": 1.26,
                 "end": 3.94,
-                "hash": "2fa6f80e0c81fb8e142f2dbbad0bceff7c21a031833b5752bc1cfd799f6b3bc6",
+                "_private_not_hashed": "0",
+                "_hash": "2fa6f80e0c81fb8e142f2dbbad0bceff7c21a031833b5752bc1cfd799f6b3bc6",
             },
             {
                 "start": 5.24,
                 "end": 6.629,
-                "hash": "34cd5da78cb163ad18996aefffcfeae864727257defc7ae68818a245ca269951",
+                "_hash": "34cd5da78cb163ad18996aefffcfeae864727257defc7ae68818a245ca269951",
             },
         ]
 
@@ -101,13 +103,13 @@ def test_get_modified_utterances(self):
                 "id": 1,
                 "start": 1.26,
                 "end": 3.94,
-                "hash": "26d514d9ce21021f51bd010d9946db0f31555ef7145067d4fe5a3b1bdcd84ce7",
+                "_hash": "26d514d9ce21021f51bd010d9946db0f31555ef7145067d4fe5a3b1bdcd84ce7",
             },
             {
                 "id": 2,
                 "start": 5.25,
                 "end": 6.629,
-                "hash": "157dc7fb355c7dc13a0ea687e9fd4a6f6c5c03526a959a64dfe1fa7562fedff4",
+                "_hash": "157dc7fb355c7dc13a0ea687e9fd4a6f6c5c03526a959a64dfe1fa7562fedff4",
             },
         ]
         dubbing = Utterance(
@@ -117,6 +119,7 @@ def test_get_modified_utterances(self):
 
         modified = dubbing.get_modified_utterances(utterances)
         assert 1 == len(modified)
+        assert 2 == modified[0]["id"]
 
     def test_get_without_empty_blocks(self):
         utterances = [
@@ -139,6 +142,7 @@ def test_get_without_empty_blocks(self):
 
         modified = dubbing.get_without_empty_blocks(utterances)
         assert 1 == len(modified)
+        assert "Hola" == modified[0]["text"]
 
     def test_add_unique_ids(self):
 
@@ -178,7 +182,7 @@ def _get_master_utterances(self):
                 "assigned_voice": "ca-ES-EnricNeural",
                 "speed": 1.0,
                 "dubbed_path": "output/jordi.central.edge.update/dubbed_chunk_1.26284375_3.94596875.mp3",
-                "hash": "b01b399ac50f80f87e704918e290ffc5ee0a1962683ba946c627124ea903480d",
+                "_hash": "b01b399ac50f80f87e704918e290ffc5ee0a1962683ba946c627124ea903480d",
             },
             {
                 "id": 2,
@@ -193,7 +197,7 @@ def _get_master_utterances(self):
                 "assigned_voice": "ca-ES-EnricNeural",
                 "speed": 1.0,
                 "dubbed_path": "output/jordi.central.edge.update/dubbed_chunk_5.24534375_6.629093750000001.mp3",
-                "hash": "629484afdecb7641e35d686d6348cee4445611690f2f77831e892d52c3128bdd",
+                "_hash": "629484afdecb7641e35d686d6348cee4445611690f2f77831e892d52c3128bdd",
             },
         ]
 
@@ -241,7 +245,7 @@ def test_update_utterances_operation_update(self):
             "assigned_voice": "ca-ES-EnricNeural",
             "speed": 1.0,
             "dubbed_path": "output/jordi.central.edge.update/dubbed_chunk_5.24534375_6.629093750000001.mp3",
-            "hash": "629484afdecb7641e35d686d6348cee4445611690f2f77831e892d52c3128bdd",
+            "_hash": "629484afdecb7641e35d686d6348cee4445611690f2f77831e892d52c3128bdd",
         }
 
     def test_load_utterances(self):
@@ -264,3 +268,46 @@ def test_load_utterances(self):
             "original_subtitles": False,
             "dubbed_subtitles": False,
         }
+
+    def _get_utterance(self):
+        return {
+            "id": 2,
+            "start": 5.24534375,
+            "end": 6.64596875,
+            "speaker_id": "SPEAKER_00",
+            "path": "output/jordi.voices/chunk_5.24534375_6.64596875.mp3",
+            "text": "I am from Barcelona.",
+            "for_dubbing": "true",
+            "gender": "Male",
+            "translated_text": "Soc de Barcelona.",
+            "assigned_voice": "2",
+            "speed": 1.0,
+            "dubbed_path": "output/jordi.voices/dubbed_chunk_5.24534375_6.64596875.mp3",
+            "_hash": "ea1d02c92026bc8cd6144a6500489333bdf0b58368817ea5116189d101c1fe9e",
+            "_assigned_voice_hash": "d4735e3a265e16eee03f59718b9b5d03019c07d8b6c51f90da3a666eec13ab35",
+            "_speaker_id_hash": "bf4f81ea701e475a4268bb9f36ddb43d7d5c0dbf4578fd5a24c3a8b5a375b4c9",
+        }
+
+    def test_get_modified_utterances_with_field_hashes(self):
+        utterances = [self._get_utterance()]
+        dubbing = Utterance(
+            target_language="cat",
+            output_directory=None,
+        )
+
+        modified = dubbing.get_modified_utterances(utterances)
+        assert 0 == len(modified)
+
+    def test_get_modified_utterance_fields_none(self):
+        utterance = self._get_utterance()
+        u = Utterance(target_language="cat", output_directory="")
+        fields = u.get_modified_utterance_fields(utterance)
+        assert len(fields) == 0
+
+    def test_get_modified_utterance_fields_speaker_id(self):
+        utterance = self._get_utterance()
+        utterance["speaker_id"] = "SPEAKER_01"
+        u = Utterance(target_language="cat", output_directory="")
+        fields = u.get_modified_utterance_fields(utterance)
+        assert len(fields) == 1
+        assert "speaker_id" == fields[0]