Audio PR - Augmentation support [PreEmphasis Filter] (#122)

* Audio Decoder PR 1 * channge image_info to sample_info to maintain a generic name for all the use-cases * Change the copyright year from 2023 to 2024 * formatting the files * Resolve PR comments * NWC * Resolve PR comments * Change decoded_img_info to decoded_video_info * Change the file_path() function to virtual from pure virtual * Minor change * Minor changes * Add the unit test file * Revert "Add the unit test file" This reverts commit e79cc06. * Fix the ROI api in python end * Fix the ROI api in python end * Remove the ROI change from pytorch.py * Add Python Unit Test for Audio * PreEmphasis Filter * Add Preemp filter augmentations files * Introduce CMake for sndfile Modify CMakeLists.txt for the same * Resolve 1st set of PR commenst * Remove commented code for last batch polices and unsued imports * ROI related changes - change from xy to wh to use for samples and channels * Fix seg fault with ROI * Remove opencv usage from the unit test * Resolve the PR comments * Remove instances of the audio_*_time - use the existing variables from Timing struct * Formatting changes in rocal_api_data_loader.cpp and add the opencl and hip conditions for audio loaders * Resolve the internal PR comments * Reformatting the file_source_reader.cpp * Remove _input_path from audio_source_evaluator and audio_read_and_decode as it is unecessary * Change the header formatting * Changes in copy_data() for audio samples * Initialize the status at the beginning * Resolve PR comments * Resolving PR comments * Resolve PR comments * Remove the unused variables * Resolve PR commensts * Cmake related changes for audio * Resolve PR comments * Add condition check to eliminate any other file extensions other than a wav file / other image formats and call open_folder deom subfolder_reading() function * Add basic testing for Audio to compare Bin files * Update audio_read_and_decode.cpp * Update rocAL_api_audio_unittest.py * Revert file source reader changes * Update master_graph.cpp * Update tensor.cpp - Remove a commented line of code * Rephrase * Introduce ROCAL_AUDIO flag Introduce flag for audio code, to be disabled when sndfile not found * Minor changes * Minor changes * Add output comparision for Audio outputs * Minor changes * Add comment * Minor changes to unit test * Remove max_frames and max_channels args * Remove max_frames, max_channels and sample rate from unit test * Remove the NONE types * Clean up unit test * Clean up code in audio python unit test * Introduce ROI in the unit test * Minor change * Format python unit test * Add python script to run audio unittests * Fix formatting issues - Minor changes * Clean up C++ audio unit test * Remove max frames and channels from decoder * Minor changes * Add Output comparison for python audio unittests * Modify rocal audio unit test Update README * Minor change * Resolve PR comments * Minor changes - Modifying the names of the arguments * Added a bried desc for the rocAL enum for border type * Add a WRN statement in PreEmphasis Filter to only use FP32 dtype * minor changes * Change the borderType enum to int32 from uint32 dtype * Minor changes * Minor change * Update README for audio unit test * Parameters for rocALAudioIterator * Minor change * Remove the reset_tensor_roi() from the PreEmphasis augmentation making sure the src and dst roi points to the same location * Del the Unit Test Files introduced earlier * Add pre_emphasis function and gollden output comparison in audio unit tests * minor change - add update val in create array * Minor variable name change * minor change * Update unit test * Minor change * minor change * Minor changes Add wav extension in file reader Add reader in unit test * Minor change * Add file reader to python audio unit test * Update C++ unit test * Update python audio unit test * Remove the unused variable output - Resolve warnings in cpp unit test * Remove the dst_roi arg passed to rpp * Minor changes Change borderType enum prefix * Name change from sample to data * Change from decoded_data_info to DecodedDataInfo * Remove audio_decoder_factory.cpp file * Minor change * Change variable name * Update the struct variable name in audio files * Minor changes * Change ROCAL_DATA_PATH to exclude rocal_data * Update ROCAL_DATA_PATH to exclude rocal_data * Use Pascal case for function names in audio decoder * Modify cmake to have SNDFILE in all capital * Minor changes * Add struct for audio info in AudioReadAndDecode * Fix merge conflict * Renaming crop_image_info to CropImageInfo * Remove - actual_host_buffers - Unused * Rename TimingDBG to TimingDbg * Move the instances of DecodedDataInfo to its base class LoaderModule * Fix a WRN msg in master_graph.cpp * Remove a dangling comment * Rename _circ_data_info to _circ_buff_data_info * Add Glob to CMakeLists.txt * Rename SndFileDecoder to GenericAudioDecoder * Fix build issues * Minor change * Update python API README.md for audio unit test * Update audio unit test README * Revert "Add Glob to CMakeLists.txt" This reverts commit 47263d9. * Fix include headers for Audio files * Fix copy data 2D * Minor changes * Pass decoded data info to load routine instead of separate vectors * Update CHANGELOG.md * Update CHANGELOG.md * Change swap_handle_time variable name in loader * Update the changelog.md * Formatting changes Add comments * Update doxygen comments * Move file source reader from readers/image to readers folder * Update README and add doxygen description * Update CMakeLists and README for audio test * Update README for audio test * Minor fix * Fix merge from PR 2 * Minor changes shard_count argument name * Rename set and get functions of data_info to decoded_data_info * Revert empty line removed in CMakeLists.txt * Removed prefix original for audio vectors * Resolve PR comments * Add @params to all args in pytorch.py * Minor changes in unit test * Minor changes * Change ROCAL instaces to rocAL in pytorch.py * Resolve the PR comments * Minor changes in decoders.py - Modify the comment for shard_size * Minor changes * Address the PR comments * Address Review comments --------- Co-authored-by: root <root@ixt-sjc2-54.local.lan> Co-authored-by: Swetha B S <swetha@multicorwareinc.com> Co-authored-by: fiona-gladwin <fionagladwin@multicorewareinc.com> Co-authored-by: Fiona-MCW <70996026+fiona-gladwin@users.noreply.github.com> Co-authored-by: Swetha B S <> Co-authored-by: SundarRajan28 <sundarrajan@multicorewareinc.com> Co-authored-by: Kiriti Gowda <kiritigowda@gmail.com>
ROCm · Jun 5, 2024 · 3fa8fa1 · 3fa8fa1
1 parent 454346c
commit 3fa8fa1
Show file tree

Hide file tree

Showing 16 changed files with 241 additions and 4 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -17,6 +17,7 @@
 * Pytorch iterator for Audio
 * Python audio unit test, and support to verify outputs
 * rocDecode for HW decode
+* Support for Audio augmentation - PreEmphasis filter
 
 ### Optimizations
 

diff --git a/rocAL/include/api/rocal_api_augmentation.h b/rocAL/include/api/rocal_api_augmentation.h
@@ -1098,4 +1098,21 @@ extern "C" RocalTensor ROCAL_API_CALL rocalSSDRandomCrop(RocalContext context, R
                                                          RocalTensorLayout output_layout = ROCAL_NONE,
                                                          RocalTensorOutputType output_datatype = ROCAL_UINT8);
 
+/*! \brief Applies preemphasis filter to the input data.
+ * \ingroup group_rocal_augmentations
+ * \param [in] context Rocal context
+ * \param [in] input Input Rocal tensor
+ * \param [in] is_output Sets to True if the output tensor is part of the graph output
+ * \param [in] preemph_coeff Preemphasis coefficient
+ * \param [in] preemph_border_type Border value policy. Possible values are "zero", "clamp", "reflect".
+ * \param [in] output_datatype The data type of the output tensor
+ * \return RocalTensor
+ */
+extern "C" RocalTensor ROCAL_API_CALL rocalPreEmphasisFilter(RocalContext context,
+                                                             RocalTensor input,
+                                                             bool is_output,
+                                                             RocalFloatParam preemph_coeff = NULL,
+                                                             RocalAudioBorderType preemph_border_type = RocalAudioBorderType::ROCAL_CLAMP,
+                                                             RocalTensorOutputType output_datatype = ROCAL_FP32);
+
 #endif  // MIVISIONX_ROCAL_API_AUGMENTATION_H
diff --git a/rocAL/include/api/rocal_api_types.h b/rocAL/include/api/rocal_api_types.h
@@ -377,6 +377,21 @@ enum RocalExternalSourceMode {
     ROCAL_EXTSOURCE_RAW_UNCOMPRESSED = 2,
 };
 
+/*! \brief rocAL Audio Border Type enum
+ * \ingroup group_rocal_types
+ */
+enum RocalAudioBorderType {
+    /*! \brief AMD ROCAL_ZERO
+     */
+    ROCAL_ZERO = 0,
+    /*! \brief AMD ROCAL_CLAMP
+     */
+    ROCAL_CLAMP = 1,
+    /*! \brief AMD ROCAL_REFLECT
+     */
+    ROCAL_REFLECT = 2
+};
+
 /*! \brief Tensor Last Batch Policies
  *  \ingroup group_rocal_types
  */

diff --git a/rocAL/include/augmentations/audio_augmentations/node_preemphasis_filter.h b/rocAL/include/augmentations/audio_augmentations/node_preemphasis_filter.h
@@ -0,0 +1,44 @@
+/*
+Copyright (c) 2024 Advanced Micro Devices, Inc. All rights reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+THE SOFTWARE.
+*/
+
+#pragma once
+#include "pipeline/graph.h"
+#include "pipeline/node.h"
+#include "parameters/parameter_factory.h"
+#include "parameters/parameter_vx.h"
+#include "rocal_api_types.h"
+
+class PreemphasisFilterNode : public Node {
+    public:
+        PreemphasisFilterNode(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs);
+        PreemphasisFilterNode() = delete;
+        void init(FloatParam *preemph_coeff, RocalAudioBorderType preemph_border);
+
+    protected:
+        void create_node() override;
+        void update_node() override;
+
+    private:
+        ParameterVX<float> _preemph_coeff;
+        constexpr static float PREEMPH_COEFF_RANGE[2] = {0.97, 0.97};
+        RocalAudioBorderType _preemph_border;
+};
diff --git a/rocAL/include/augmentations/augmentations_nodes.h b/rocAL/include/augmentations/augmentations_nodes.h
@@ -55,3 +55,4 @@ THE SOFTWARE.
 #include "augmentations/node_copy.h"
 #include "augmentations/node_nop.h"
 #include "augmentations/node_sequence_rearrange.h"
+#include "augmentations/audio_augmentations/node_preemphasis_filter.h"
diff --git a/rocAL/source/api/rocal_api_augmentation.cpp b/rocAL/source/api/rocal_api_augmentation.cpp
@@ -2155,3 +2155,35 @@ rocalNop(
     }
     return output;
 }
+
+RocalTensor ROCAL_API_CALL
+rocalPreEmphasisFilter(RocalContext p_context,
+                       RocalTensor p_input,
+                       bool is_output,
+                       RocalFloatParam p_preemph_coeff,
+                       RocalAudioBorderType preemph_border_type,
+                       RocalTensorOutputType output_datatype) {
+    Tensor* output = nullptr;
+    if ((p_context == nullptr) || (p_input == nullptr)) {
+        ERR("Invalid ROCAL context or invalid input tensor")
+        return output;
+    }
+    auto context = static_cast<Context*>(p_context);
+    auto input = static_cast<Tensor*>(p_input);
+    auto preemph_coeff = static_cast<FloatParam*>(p_preemph_coeff);
+    try {
+        RocalTensorDataType op_tensor_datatype = static_cast<RocalTensorDataType>(output_datatype);
+        if (op_tensor_datatype != RocalTensorDataType::FP32) {
+            WRN("Only FP32 dtype is supported for PreEmphasis filter augmentation.")
+            op_tensor_datatype = RocalTensorDataType::FP32;
+        }
+        TensorInfo output_info = input->info();
+        output_info.set_data_type(op_tensor_datatype);
+        output = context->master_graph->create_tensor(output_info, is_output);
+        context->master_graph->add_node<PreemphasisFilterNode>({input}, {output})->init(preemph_coeff, preemph_border_type);
+    } catch (const std::exception& e) {
+        context->capture_error(e.what());
+        ERR(e.what())
+    }
+    return output;
+}
diff --git a/rocAL/source/augmentations/audio_augmentations/node_preemphasis_filter.cpp b/rocAL/source/augmentations/audio_augmentations/node_preemphasis_filter.cpp
@@ -0,0 +1,50 @@
+/*
+Copyright (c) 2024 Advanced Micro Devices, Inc. All rights reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+THE SOFTWARE.
+*/
+
+#include <vx_ext_rpp.h>
+#include "augmentations/audio_augmentations/node_preemphasis_filter.h"
+#include "pipeline/exception.h"
+
+PreemphasisFilterNode::PreemphasisFilterNode(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) : Node(inputs, outputs),
+                                                                                                                          _preemph_coeff(PREEMPH_COEFF_RANGE[0], PREEMPH_COEFF_RANGE[1]) {}
+
+void PreemphasisFilterNode::create_node() {
+    if (_node)
+        return;
+    _preemph_coeff.create_array(_graph, VX_TYPE_FLOAT32, _batch_size);
+    vx_scalar border_type_vx = vxCreateScalar(vxGetContext((vx_reference)_graph->get()), VX_TYPE_INT32, &_preemph_border);
+    _node = vxExtRppPreemphasisFilter(_graph->get(), _inputs[0]->handle(), _inputs[0]->get_roi_tensor(), _outputs[0]->handle(), _preemph_coeff.default_array(), border_type_vx);
+    vx_status status;
+    if ((status = vxGetStatus((vx_reference)_node)) != VX_SUCCESS)
+        THROW("Adding the vxExtRppPreemphasisFilter node failed: " + TOSTR(status))
+}
+
+void PreemphasisFilterNode::update_node() {
+    _preemph_coeff.update_array();
+}
+
+void PreemphasisFilterNode::init(FloatParam *preemph_coeff, RocalAudioBorderType preemph_border) {
+    if (preemph_coeff == nullptr)
+        ERR("Invalid pre-Emphasis co-efficient passed")
+    _preemph_coeff.set_param(core(preemph_coeff));
+    _preemph_border = preemph_border;
+}
diff --git a/rocAL_pybind/amd/rocal/fn.py b/rocAL_pybind/amd/rocal/fn.py
@@ -1056,7 +1056,17 @@ def box_iou_matcher(*inputs, anchors, high_threshold=0.5,
     return (box_iou_matcher, [])
 
 
-def external_source(*inputs, source, device=None, color_format=types.RGB, random_shuffle=False, mode=types.EXTSOURCE_FNAME, max_width=2000, max_height=2000):
+def external_source(source, device=None, color_format=types.RGB, random_shuffle=False, mode=types.EXTSOURCE_FNAME, max_width=2000, max_height=2000):
+    """
+    External Source Reader - User can pass a iterator or callable source.
+    @param source (iterator or callable)                                 The source iterator or callable object.
+    @param device (string, optional, default = None)                     Parameter unused for augmentation
+    @color_format (type, optional, default = RGB)                        Tensor color format for the reader. Default is RGB format.
+    @random_shuffle (bool, optional, default= False)                     Data would be randomly shuffled if set to True.
+    @mode (type, optional, default = EXTSOURCE_FNAME)                    The Default mode would be External Source File Name. The External Source Mode can be FileName, Raw Compressed, Raw Uncompressed.
+    @max_width (int, optional, default = 2000)                           The Max Width to which the source images would be decoded to.
+    @max_height (int, optional, default = 2000)                          The Max Height to which the source images would be decoded to.
+    """
     # pybind call arguments
     Pipeline._current_pipeline._is_external_source_operator = True
     Pipeline._current_pipeline._external_source = iter(source)
@@ -1068,3 +1078,27 @@ def external_source(*inputs, source, device=None, color_format=types.RGB, random
     external_source_operator = b.externalFileSource(
         Pipeline._current_pipeline._handle, *(kwargs_pybind.values()))
     return (external_source_operator, [])  # Labels is Empty
+
+def preemphasis_filter(*inputs, border=types.CLAMP, preemph_coeff=0.97, output_dtype=types.FLOAT):
+    """
+    Applies preemphasis filter to the input data.
+    This filter, in simple form, can be expressed by the formula:
+    Y[t] = X[t] - coeff * X[t-1]    if t > 1
+    Y[t] = X[t] - coeff * X_border  if t == 0
+    with X and Y being the input and output signal, respectively.
+    The value of X_border depends on the border argument:
+    X_border = 0                    if border_type == 'zero'
+    X_border = X[0]                 if border_type == 'clamp'
+    X_border = X[1]                 if border_type == 'reflect'
+
+    @param inputs (list)                                                 The input sample to which preEmphasisFilter is applied.
+    @param border                                                        The border value policy. The possible values are "CLAMP", "ZERO", "REFLECT"
+    @param preemph_coeff (float , optional, default = 0.97)              The preEmphasisFilter co-efficient.
+    @output_dtype (type, optional, default = types.FLOAT)                Tensor dtype for the augmentation output. Default is types.FLOAT.
+    """
+    preemph_coeff_float_param = b.createFloatParameter(preemph_coeff)
+    kwargs_pybind = {"input_audio0": inputs[0], "is_output": False,
+                    "preemph_coeff": preemph_coeff_float_param, "preemph_border_type": border,
+                    "output_dtype" :output_dtype}
+    preemphasis_output = b.preEmphasisFilter(Pipeline._current_pipeline._handle ,*(kwargs_pybind.values()))
+    return (preemphasis_output)
diff --git a/rocAL_pybind/amd/rocal/types.py b/rocAL_pybind/amd/rocal/types.py
@@ -94,6 +94,11 @@
 from rocal_pybind.types import EXTSOURCE_RAW_COMPRESSED
 from rocal_pybind.types import EXTSOURCE_RAW_UNCOMPRESSED
 
+#     RocalAudioBorderType
+from rocal_pybind.types import ZERO
+from rocal_pybind.types import CLAMP
+from rocal_pybind.types import REFLECT
+
 #     RocalLastBatchPolicy
 from rocal_pybind.types import LAST_BATCH_FILL
 from rocal_pybind.types import LAST_BATCH_DROP
@@ -158,6 +163,10 @@
     EXTSOURCE_RAW_COMPRESSED: ("EXTSOURCE_RAW_COMPRESSED", EXTSOURCE_RAW_COMPRESSED),
     EXTSOURCE_RAW_UNCOMPRESSED: ("EXTSOURCE_RAW_UNCOMPRESSED", EXTSOURCE_RAW_UNCOMPRESSED),
 
+    ZERO: ("ZERO", ZERO),
+    CLAMP: ("CLAMP", CLAMP),
+    REFLECT: ("REFLECT", REFLECT),
+
     LAST_BATCH_FILL : ("LAST_BATCH_FILL", LAST_BATCH_FILL),
     LAST_BATCH_DROP : ("LAST_BATCH_DROP", LAST_BATCH_DROP),
     LAST_BATCH_PARTIAL : ("LAST_BATCH_PARTIAL", LAST_BATCH_PARTIAL),

diff --git a/rocAL_pybind/rocal_pybind.cpp b/rocAL_pybind/rocal_pybind.cpp
@@ -396,6 +396,11 @@ PYBIND11_MODULE(rocal_pybind, m) {
         .value("EXTSOURCE_RAW_COMPRESSED", ROCAL_EXTSOURCE_RAW_COMPRESSED)
         .value("EXTSOURCE_RAW_UNCOMPRESSED", ROCAL_EXTSOURCE_RAW_UNCOMPRESSED)
         .export_values();
+    py::enum_<RocalAudioBorderType>(types_m,"RocalAudioBorderType", "Rocal Audio Border Type")
+        .value("ZERO", ROCAL_ZERO)
+        .value("CLAMP", ROCAL_CLAMP)
+        .value("REFLECT", ROCAL_REFLECT)
+        .export_values();
     py::enum_<RocalLastBatchPolicy>(types_m, "RocalLastBatchPolicy", "Rocal Last Batch Policy")
         .value("LAST_BATCH_FILL", ROCAL_LAST_BATCH_FILL)
         .value("LAST_BATCH_DROP", ROCAL_LAST_BATCH_DROP)
@@ -716,5 +721,7 @@ PYBIND11_MODULE(rocal_pybind, m) {
           py::return_value_policy::reference);
     m.def("lensCorrection", &rocalLensCorrection,
           py::return_value_policy::reference);
+    m.def("preEmphasisFilter", &rocalPreEmphasisFilter, 
+          py::return_value_policy::reference);
 }
 }  // namespace rocal
diff --git a/tests/cpp_api/audio_tests/README.md b/tests/cpp_api/audio_tests/README.md
@@ -38,3 +38,4 @@ python3 audio_tests.py --gpu <0/1> --downmix <True/False> --test_case <case_numb
 
 **Available Test Cases**
 * Case 0 - Audio Decoder
+* Case 1 - PreEmphasis Filter
diff --git a/tests/cpp_api/audio_tests/audio_tests.cpp b/tests/cpp_api/audio_tests/audio_tests.cpp
@@ -149,6 +149,15 @@ int test(int test_case, const char *path, int qa_mode, int downmix, int gpu) {
             case_name = "audio_decoder";
             std::cout << "Running AUDIO DECODER" << std::endl;
         } break;
+        case 1: {
+            std::cout << "Running PREEMPHASIS" << std::endl;
+            case_name = "preemphasis_filter";
+            RocalTensorOutputType tensorOutputType = RocalTensorOutputType::ROCAL_FP32;
+            RocalAudioBorderType preemph_border_type = RocalAudioBorderType::ROCAL_CLAMP;
+            RocalFloatParam p_preemph_coeff = rocalCreateFloatParameter(0.97);
+            rocalPreEmphasisFilter(handle, decoded_output, true, p_preemph_coeff, preemph_border_type, tensorOutputType);
+
+        } break;
         default: {
             std::cout << "Not a valid test case ! Exiting!\n";
             return -1;

diff --git a/tests/cpp_api/audio_tests/audio_tests.py b/tests/cpp_api/audio_tests/audio_tests.py
@@ -25,7 +25,8 @@
 import argparse
 
 test_case_augmentation_map = {
-    0: "audio_decoder"
+    0: "audio_decoder",
+    1: "preemphasis_filter"
 }
 
 def run_unit_test(src_path, qa_mode, gpu, downmix, build_folder_path, case_list):

diff --git a/tests/python_api/README.md b/tests/python_api/README.md
@@ -139,4 +139,5 @@ python3 audio_unit_test.py --audio_path=<path_to_data> --test_case <case(0-1)> -
 ```
 
 **Available Test Cases**
-* Case 0 - Audio Decoder
+* Case 0 - Audio Decoder
+* Case 1 - PreEmphasis Filter
diff --git a/tests/python_api/audio_unit_test.py b/tests/python_api/audio_unit_test.py
@@ -35,6 +35,7 @@
 
 test_case_augmentation_map = {
     0: "audio_decoder",
+    1: "preemphasis_filter"
 }
 
 def plot_audio_wav(audio_tensor, idx):
@@ -80,6 +81,18 @@ def audio_decoder_pipeline(path):
         num_shards=1,
         stick_to_shard=False)
 
+@pipeline_def(seed=seed)
+def pre_emphasis_filter_pipeline(path):
+    audio, labels = fn.readers.file(file_root=path)
+    decoded_audio = fn.decoders.audio(
+        audio,
+        file_root=path,
+        downmix=False,
+        shard_id=0,
+        num_shards=1,
+        stick_to_shard=False)
+    return fn.preemphasis_filter(decoded_audio)
+
 def main():
     args = parse_args()
 
@@ -131,6 +144,8 @@ def main():
         case_name = test_case_augmentation_map.get(case)
         if case_name == "audio_decoder":
             audio_pipeline = audio_decoder_pipeline(batch_size=batch_size, num_threads=num_threads, device_id=device_id, rocal_cpu=rocal_cpu, path=audio_path)
+        if case_name == "preemphasis_filter":
+            audio_pipeline = pre_emphasis_filter_pipeline(batch_size=batch_size, num_threads=num_threads, device_id=device_id, rocal_cpu=rocal_cpu, path=audio_path)
         audio_pipeline.build()
         audio_loader = ROCALAudioIterator(audio_pipeline, auto_reset=True)
         cnt = 0

diff --git a/tests/python_api/parse_config.py b/tests/python_api/parse_config.py
@@ -111,7 +111,7 @@ def parse_args():
         'audio-python-unittest', 'audio-python-unittest-related options')
     audio_unit_test.add_argument('--audio_path', type=str, default="",
                                   help='audio files path')
-    audio_unit_test.add_argument('--test_case', type=int, default=0,
+    audio_unit_test.add_argument('--test_case', type=int, default=None,
                                   help='test case')
     audio_unit_test.add_argument('--qa_mode', type=int, default=1,
                                   help='enable qa mode to compare audio output with ref outputs')