update to aic-sdk 2.0

aic_sdk.py

@@ -1,45 +0,0 @@
-import numpy as np
-from dotenv import load_dotenv
-from aic import Model, AICModelType, AICParameter
-import librosa
-import soundfile as sf
-
-load_dotenv()
-
-
-class SDKParams:
-    def __init__(self, enhancement_level: float, sdk_key: str):
-        self.enhancement_level = enhancement_level
-        self.sdk_key = sdk_key
-
-
-def process_file_sdk(input_path: str, output_path: str, sdk_params: SDKParams):
-    # Load audio
-    audio, sample_rate = librosa.load(input_path, sr=48000, mono=True)
-    audio = audio.reshape(1, -1)  # Convert to planar format
-    output = np.zeros_like(audio)
-    load_dotenv()
-    with Model(
-        AICModelType.QUAIL_L,
-        license_key=sdk_params.sdk_key,
-        sample_rate=48000,
-        channels=1,
-    ) as model:
-        model.set_parameter(AICParameter.ENHANCEMENT_LEVEL, sdk_params.enhancement_level)
-        # Process in chunks
-        chunk_size = model.optimal_num_frames()
-        for i in range(0, audio.shape[1], chunk_size):
-            chunk = audio[:, i : i + chunk_size]
-            # Pad last chunk if needed
-            if chunk.shape[1] < chunk_size:
-                last_chunk_size = chunk.shape[1]
-                padded = np.zeros((1, chunk_size), dtype=audio.dtype)
-                padded[:, : chunk.shape[1]] = chunk
-                chunk = padded
-                enhanced_chunk = model.process(chunk)
-                output[:, i : i + last_chunk_size] = enhanced_chunk[:, :last_chunk_size]
-                break
-            enhanced_chunk = model.process(chunk)
-            output[:, i : i + chunk_size] = enhanced_chunk[:, :chunk_size]
-    # Save result
-    sf.write(output_path, output.T, sample_rate)

app.py

@@ -16,8 +16,7 @@ from constants import (
     MINUTES_KEEP,
     EXAMPLES,
 )
-from aic_api import ApiParamsV2, process_file_v2
-from aic_sdk import SDKParams, process_file_sdk
+from sdk import SDKParams, process_file_sdk
 from audio_tools import spec_image, mix_at_snr
 import shutil
 import tempfile

requirements.txt

@@ -4,6 +4,6 @@ soundfile>=0.12.1
 aiohttp>=3.9,<4
 librosa>=0.10.1,<0.11
 loguru~=0.7
-aic-sdk
+aic-sdk>=2.0.0
 dotenv
 resampy

sdk.py

@@ -0,0 +1,52 @@
+from platform import processor
+
+import numpy as np
+from dotenv import load_dotenv
+import aic_sdk as aic
+import librosa
+import soundfile as sf
+
+load_dotenv()
+
+
+
+
+
+
+class SDKParams:
+    def __init__(self, enhancement_level: float, sdk_key: str, model_id: str = "quail-vf-1.1-l-16khz"):
+        self.enhancement_level = enhancement_level
+        self.sdk_key = sdk_key
+        self.model_path=  aic.Model.download(model_id, "./models")
+        
+
+def process_file_sdk(input_path: str, output_path: str, sdk_params: SDKParams):
+    # Load audio
+    audio, sr = librosa.load(input_path, sr=48000, mono=True)
+    audio = audio.reshape(1, -1)  # Convert to planar format
+    output = np.zeros_like(audio)
+    load_dotenv()
+    model = aic.Model.from_file(sdk_params.model_path)
+    optimal_frames = model.get_optimal_num_frames(48000)
+    config = aic.ProcessorConfig(sample_rate=sr, num_channels=1, num_frames= optimal_frames,allow_variable_frames=False )
+    processor = aic.Processor(model, sdk_params.sdk_key, config)
+    proc_ctx = processor.get_processor_context()
+    #proc_ctx.set_parameter(aic.ProcessorParameter.EnhancementLevel, sdk_params.enhancement_level)
+    #proc_ctx.set_parameter(aic.ProcessorParameter.VoiceGain, 1.5)
+    #proc_ctx.set_parameter(aic.ProcessorParameter.Bypass, 0.0)
+    chunk_size = optimal_frames
+    for i in range(0, audio.shape[1], chunk_size):
+        chunk = audio[:, i : i + chunk_size]
+        # Pad last chunk if needed
+        if chunk.shape[1] < chunk_size:
+            last_chunk_size = chunk.shape[1]
+            padded = np.zeros((1, chunk_size), dtype=audio.dtype)
+            padded[:, : chunk.shape[1]] = chunk
+            chunk = padded
+            enhanced_chunk = processor.process(chunk)
+            output[:, i : i + last_chunk_size] = enhanced_chunk[:, :last_chunk_size]
+            break
+        enhanced_chunk = processor.process(chunk)
+        output[:, i : i + chunk_size] = enhanced_chunk[:, :chunk_size]
+    # Save result
+    sf.write(output_path, output.T, sr)