sonusai 0.15.3__tar.gz → 0.15.5__tar.gz

{sonusai-0.15.3 → sonusai-0.15.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: sonusai
-Version: 0.15.3
+Version: 0.15.5
 Summary: Framework for building deep neural network models for sound, speech, and voice AI
 Home-page: https://aaware.com
 License: GPL-3.0-only
@@ -17,12 +17,12 @@ Classifier: Programming Language :: Python :: 3.11
 Requires-Dist: PyYAML (>=6.0.1,<7.0.0)
 Requires-Dist: aixplain (>=0.2.6,<0.3.0)
 Requires-Dist: dataclasses-json (>=0.6.1,<0.7.0)
-Requires-Dist: deepgram-sdk (>=2.11.0,<3.0.0)
+Requires-Dist: deepgram-sdk (>=3.0.0,<4.0.0)
 Requires-Dist: docopt (>=0.6.2,<0.7.0)
 Requires-Dist: faster-whisper (>=0.10.0,<0.11.0)
 Requires-Dist: ffmpeg (>=1.4,<2.0)
 Requires-Dist: greenlet (>=3.0.1,<4.0.0)
-Requires-Dist: grpcio (==1.59.2)
+Requires-Dist: grpcio (==1.60.0)
 Requires-Dist: jiwer (>=3.0.3,<4.0.0)
 Requires-Dist: keras-tuner (>=1.4.5,<2.0.0)
 Requires-Dist: librosa (>=0.10.1,<0.11.0)
@@ -35,7 +35,7 @@ Requires-Dist: pandas (>=2.1.1,<3.0.0)
 Requires-Dist: pesq (>=0.0.4,<0.0.5)
 Requires-Dist: pyaaware (>=1.5.3,<2.0.0)
 Requires-Dist: pydub (>=0.25.1,<0.26.0)
-Requires-Dist: pystoi (>=0.3.3,<0.4.0)
+Requires-Dist: pystoi (>=0.4.0,<0.5.0)
 Requires-Dist: python-magic (>=0.4.27,<0.5.0)
 Requires-Dist: requests (>=2.31.0,<3.0.0)
 Requires-Dist: resampy (>=0.4.2,<0.5.0)
@@ -45,7 +45,7 @@ Requires-Dist: sh (>=2.0.6,<3.0.0)
 Requires-Dist: soundfile (>=0.12.1,<0.13.0)
 Requires-Dist: sox (>=1.4.1,<2.0.0)
 Requires-Dist: speechrecognition (>=3.10.0,<4.0.0)
-Requires-Dist: tensorflow (>=2.14.0,<3.0.0)
+Requires-Dist: tensorflow (>=2.15.0,<3.0.0)
 Requires-Dist: tensorflow-addons (>=0.23.0,<0.24.0)
 Requires-Dist: tf2onnx (>=1.15.1,<2.0.0)
 Requires-Dist: torch (>=2.1.0,<3.0.0)

{sonusai-0.15.3 → sonusai-0.15.5}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "sonusai"
-version = "0.15.3"
+version = "0.15.5"
 description = "Framework for building deep neural network models for sound, speech, and voice AI"
 authors = ["Chris Eddington <chris@aaware.com>", "Jason Calderwood <jason@aaware.com>"]
 maintainers = ["Chris Eddington <chris@aaware.com>", "Jason Calderwood <jason@aaware.com>"]
@@ -17,20 +17,20 @@ aawscd_probwrite = 'sonusai.aawscd_probwrite:main'
 PyYAML = "^6.0.1"
 aixplain = "^0.2.6"
 dataclasses-json = "^0.6.1"
-deepgram-sdk = "^2.11.0"
+deepgram-sdk = "^3.0.0"
 docopt = "^0.6.2"
 #einops = "^0"
 faster-whisper = "^0.10.0"
 ffmpeg = "^1.4"
 greenlet = "^3.0.1"
-grpcio = "1.59.2"
+grpcio = "1.60.0"
 jiwer = "^3.0.3"
 keras-tuner = "^1.4.5"
 librosa = "^0.10.1"
 lightning = "^2.1.0"
 matplotlib = "^3.8.0"
 onnx = "1.14.1"
-#onnxruntime-gpu = "^1"
+#onnxruntime-gpu = "^1.16.1"
 onnxruntime = "^1.16.1"
 #openai-whisper = "^20231117"
 paho-mqtt = "^1.6.1"
@@ -38,7 +38,7 @@ pandas = "^2.1.1"
 pesq = "^0.0.4"
 pyaaware = "^1.5.3"
 pydub = "^0.25.1"
-pystoi = "^0.3.3"
+pystoi = "^0.4.0"
 python = ">=3.9,<3.12"
 python-magic = "^0.4.27"
 requests = "^2.31.0"
@@ -49,7 +49,7 @@ sh = "^2.0.6"
 soundfile = "^0.12.1"
 sox = "^1.4.1"
 speechrecognition = "^3.10.0"
-tensorflow = "^2.14.0"
+tensorflow = "^2.15.0"
 tensorflow-addons = "^0.23.0"
 tf2onnx = "^1.15.1"
 torch = "^2.1.0"

{sonusai-0.15.3 → sonusai-0.15.5}/sonusai/mixture/augmentation.py

@@ -58,11 +58,14 @@ def _expand_rules(expanded_rules: list[dict], rule: dict) -> list[dict]:
             if not eq_rule_is_valid(rule[key]):
                 raise SonusAIError(f'Invalid augmentation value for {key}: {rule[key]}')
 
-            if all(isinstance(el, list) for el in rule[key]):
+            if all(isinstance(el, list) or (isinstance(el, str) and el == 'none') for el in rule[key]):
                 # Expand multiple rules
                 for value in rule[key]:
                     expanded_rule = deepcopy(rule)
-                    expanded_rule[key] = deepcopy(value)
+                    if isinstance(value, str) and value == 'none':
+                        expanded_rule[key] = None
+                    else:
+                        expanded_rule[key] = deepcopy(value)
                     _expand_rules(expanded_rules, expanded_rule)
                 return expanded_rules
 
@@ -80,13 +83,31 @@ def _expand_rules(expanded_rules: list[dict], rule: dict) -> list[dict]:
                 return expanded_rules
             else:
                 rule[key] = convert_string_to_number(rule[key])
-                if not (isinstance(rule[key], float | int) or rule[key].startswith('rand')):
+                if not (isinstance(rule[key], float | int) or
+                        rule[key].startswith('rand') or
+                        rule[key] == 'none'):
                     raise SonusAIError(f'Invalid augmentation value for {key}: {rule[key]}')
 
     expanded_rules.append(rule)
     return expanded_rules
 
 
+def _generate_none_rule(rule: dict) -> dict:
+    """Generate a new rule from a rule that contains 'none' directives
+
+    :param rule: Rule
+    :return: New rule
+    """
+    from copy import deepcopy
+
+    out_rule = deepcopy(rule)
+    for key in out_rule:
+        if out_rule[key] == 'none':
+            out_rule[key] = None
+
+    return out_rule
+
+
 def _generate_random_rule(rule: dict, num_ir: int = 0) -> dict:
     """Generate a new rule from a rule that contains 'rand' directives
 
@@ -238,7 +259,7 @@ def _parse_ir(rule: dict, num_ir: int) -> dict:
     from .helpers import generic_ids_to_list
 
     def _resolve_str(rule_in: str) -> str | list[int]:
-        if rule_in == 'rand':
+        if rule_in in ['rand', 'none']:
             return rule_in
 
         rule_out = generic_ids_to_list(num_ir, rule_in)
@@ -308,6 +329,7 @@ def augmentation_from_rule(rule: AugmentationRule, num_ir: int) -> Augmentation:
 
     processed_rule = rule.to_dict()
     del processed_rule['mixup']
+    processed_rule = _generate_none_rule(processed_rule)
     if _rule_has_rand(processed_rule):
         processed_rule = _generate_random_rule(processed_rule, num_ir)
 

{sonusai-0.15.3 → sonusai-0.15.5}/sonusai/mixture/eq_rule_is_valid.py

@@ -7,13 +7,16 @@ def eq_rule_is_valid(rule: Any) -> bool:
     An EQ rule must be a tuple of length 3 or a list of length 3 tuples.
     """
 
-    # Must be a list
+    # Must be a list or string equal to 'none'
+    if isinstance(rule, str) and rule == 'none':
+        return True
+
     if not isinstance(rule, list):
         return False
 
     if len(rule) != 3:
         # If the length is not 3, then all elements must also be lists
-        if not all(isinstance(el, list) and len(el) == 3 for el in rule):
+        if not all(_check_for_none(el) for el in rule):
             return False
         rules = rule
     else:
@@ -24,9 +27,22 @@ def eq_rule_is_valid(rule: Any) -> bool:
         if not all(isinstance(el, float | int | str) for el in r):
             return False
 
+        if isinstance(r, str) and r == 'none':
+            continue
+
         for el in r:
             # If a string, item must start with 'rand'
             if isinstance(el, str) and not el.startswith('rand'):
                 return False
 
     return True
+
+
+def _check_for_none(rule: Any) -> bool:
+    """Check if EQ rule is 'none'
+    """
+    if isinstance(rule, str) and rule == 'none':
+        return True
+    if isinstance(rule, list) and len(rule) == 3:
+        return True
+    return False

{sonusai-0.15.3 → sonusai-0.15.5}/sonusai/onnx_predict.py

@@ -223,6 +223,9 @@ def pad_and_predict(feature: Feature,
             model = rt.InferenceSession(model_name, providers=['CPUExecutionProvider'])
 
     predict_arr = np.vstack(predict)
+    # Combine [sequences, batch_size, ...] into [frames, ...]
+    predict_shape = predict_arr.shape
+    predict_arr = np.reshape(predict_arr, [predict_shape[0] * predict_shape[1], *predict_shape[2:]])
     predict_arr, _ = reshape_outputs(predict=predict_arr, timesteps=timesteps)
     predict_arr = predict_arr[:frames, :]