feat: Add MelWeightMatrix operators

docs/framework/operators/tensor/tensor.mel_weight_matrix.md

@@ -0,0 +1,51 @@
+# TensorTrait::mel_weight_matrix
+
+```rust
+        fn mel_weight_matrix(num_mel_bins: usize, dft_length: usize, sample_rate: usize, lower_edge_hertz: T, upper_edge_hertz: T) -> Tensor<T>;
+```
+
+Generate a MelWeightMatrix that can be used to re-weight a Tensor containing a linearly sampled frequency spectra (from DFT or STFT) into num_mel_bins frequency information based on the [lower_edge_hertz, upper_edge_hertz] range on the mel scale. 
+This function defines the mel scale in terms of a frequency in hertz according to the following formula:
+```
+mel(f) = 2595 * log10(1 + f/700)
+```
+In the returned matrix, all the triangles (filterbanks) have a peak value of 1.0.
+The returned MelWeightMatrix can be used to right-multiply a spectrogram S of shape [frames, num_spectrogram_bins] of linear scale spectrum values (e.g. STFT magnitudes) to generate a “mel spectrogram” M of shape [frames, num_mel_bins].
+## Args
+
+* `num_mel_bins `(`usize`) - The number of bands in the mel spectrum.
+* `dft_length `(`usize`) - The size of the original DFT. The size of the original DFT is used to infer the size of the onesided DFT, which is understood to be floor(dft_length/2) + 1, i.e. the spectrogram only contains the nonredundant DFT bins.
+* `sample_rate `(`usize`) - Samples per second of the input signal used to create the spectrogram. Used to figure out the frequencies corresponding to each spectrogram bin, which dictates how they are mapped into the mel scale.
+* `lower_edge_hertz `(T) - Lower bound on the frequencies to be included in the mel spectrum. This corresponds to the lower edge of the lowest triangular band.
+* `upper_edge_hertz  `(T) - The desired top edge of the highest frequency band.
+
+## Returns
+
+* A `Tensor<T>` The Mel Weight Matrix. The output has the shape: [floor(dft_length/2) + 1][num_mel_bins].
+
+## Examples
+
+```rust
+use orion::operators::tensor::{FP16x16Tensor, FP16x16TensorAdd};
+use core::array::{ArrayTrait, SpanTrait};
+use orion::operators::tensor::FP16x16TensorPartialEq;
+use orion::utils::{assert_eq, assert_seq_eq};
+use orion::operators::tensor::{TensorTrait, Tensor};
+use orion::numbers::{FixedTrait, FP16x16};
+
+
+fn example() -> Tensor<FP16x16> {
+    return TensorTrait::mel_weight_matrix(8, 16, 8192, FP16x16 { mag: 0, sign: false }, FP16x16 { mag: 268435456, sign: false });
+}
+>>> [
+        [65536, 65536, 0, 0, 0, 0, 0, 0],
+        [0, 0, 65536, 65536, 0, 0, 0, 0],   
+        [0, 0, 0, 0, 65536, 0, 0, 0],
+        [0, 0, 0, 0, 0, 65536, 0, 0],
+        [0, 0, 0, 0, 0, 0, 65536, 0],   
+        [0, 0, 0, 0, 0, 0, 0, 65536],
+        [0, 0, 0, 0, 0, 0, 0, 0],
+        [0, 0, 0, 0, 0, 0, 0, 0],
+        [0, 0, 0, 0, 0, 0, 0, 0],
+    ]
+```

nodegen/node/mel_weight_matrix.py

@@ -0,0 +1,97 @@
+import numpy as np
+from nodegen.node import RunAll
+from ..helpers import make_test, to_fp, Tensor, Dtype, FixedImpl, Trait, get_data_statement
+
+def mel_weight_matrix(num_mel_bins, dft_length, sample_rate, lower_edge_hertz, upper_edge_hertz) -> np.ndarray:  # type: ignore
+    # num_mel_bins = np.int32(8)
+    # dft_length = np.int32(16)
+    # sample_rate = np.int32(8192)
+    # lower_edge_hertz = np.float32(0)
+    # upper_edge_hertz = np.float32(8192 / 2)
+    num_spectrogram_bins = dft_length // 2 + 1
+    frequency_bins = np.arange(0, num_mel_bins + 2)
+
+    low_frequency_mel = 2595 * np.log10(1 + lower_edge_hertz / 700)
+    high_frequency_mel = 2595 * np.log10(1 + upper_edge_hertz / 700)
+    mel_step = (high_frequency_mel - low_frequency_mel) / frequency_bins.shape[0]
+
+    frequency_bins = frequency_bins * mel_step + low_frequency_mel
+    frequency_bins = 700 * (np.power(10, (frequency_bins / 2595)) - 1)
+    frequency_bins = ((dft_length + 1) * frequency_bins) // sample_rate
+    frequency_bins = frequency_bins.astype(int)
+
+    output = np.zeros((num_spectrogram_bins, num_mel_bins))
+    output.flags.writeable = True
+
+    for i in range(num_mel_bins):
+        lower_frequency_value = frequency_bins[i]  # left
+        center_frequency_point = frequency_bins[i + 1]  # center
+        higher_frequency_point = frequency_bins[i + 2]  # right
+        low_to_center = center_frequency_point - lower_frequency_value
+        if low_to_center == 0:
+            output[center_frequency_point, i] = 1
+        else:
+            for j in range(lower_frequency_value, center_frequency_point + 1):
+                output[j, i] = float(j - lower_frequency_value) / float(
+                    low_to_center
+                )
+        center_to_high = higher_frequency_point - center_frequency_point
+        if center_to_high > 0:
+            for j in range(center_frequency_point, higher_frequency_point):
+                output[j, i] = float(higher_frequency_point - j) / float(
+                    center_to_high
+                )
+    return output
+
+
+class Mel_weight_matrix(RunAll):
+
+    # @staticmethod
+    # # We test here with u32 implementation.
+    # def u32():
+    #     num_mel_bins = np.int32(8)
+    #     dft_length = np.int32(16)
+    #     sample_rate = np.int32(256)
+    #     lower_edge_hertz = np.int32(0)
+    #     upper_edge_hertz = np.int32(256 / 2)
+    #     args_str = get_data_statement(np.array([lower_edge_hertz, upper_edge_hertz]), Dtype.U32)
+    #     y = mel_weight_matrix(num_mel_bins, dft_length, sample_rate, lower_edge_hertz, upper_edge_hertz)
+    #     print(y)
+
+    #     # Convert the floats values in `y` to fixed points with `to_fp` method:
+    #     y = Tensor(Dtype.U32, y.shape, y.flatten())
+
+    #     # Define the name of the generated folder. 
+    #     name = "mel_weight_matrix_u32"
+    #     # Invoke `make_test` method to generate corresponding Cairo tests:
+    #     make_test(
+    #         [], # List of input tensors.
+    #         y, # The expected output result.
+    #         f"TensorTrait::mel_weight_matrix({f'{num_mel_bins}, {dft_length}, {sample_rate}, '+', '.join(args_str)})", # The code signature.
+    #         name # The name of the generated folder.
+    #     )
+
+    @staticmethod
+    # We test here with fp16x16 implementation.
+    def fp16x16():
+        num_mel_bins = np.int32(8)
+        dft_length = np.int32(16)
+        sample_rate = np.int32(8192)
+        lower_edge_hertz = np.float32(0)
+        upper_edge_hertz = np.float32(8192 / 2)
+        args_str = get_data_statement(to_fp(np.array([lower_edge_hertz, upper_edge_hertz]).flatten(), FixedImpl.FP16x16), Dtype.FP16x16)
+        y = mel_weight_matrix(num_mel_bins, dft_length, sample_rate, lower_edge_hertz, upper_edge_hertz)
+        print(y)
+
+        # Convert the floats values in `y` to fixed points with `to_fp` method:
+        y = Tensor(Dtype.FP16x16, y.shape, to_fp(y.flatten(), FixedImpl.FP16x16))
+
+        # Define the name of the generated folder. 
+        name = "mel_weight_matrix_fp16x16"
+        # Invoke `make_test` method to generate corresponding Cairo tests:
+        make_test(
+            [], # List of input tensors.
+            y, # The expected output result.
+            f"TensorTrait::mel_weight_matrix({f'{num_mel_bins}, {dft_length}, {sample_rate}, '+', '.join(args_str)})", # The code signature.
+            name # The name of the generated folder.
+        )

src/operators/tensor/core.cairo

@@ -131,6 +131,7 @@ impl TensorSerde<T, impl TSerde: Serde<T>, impl TDrop: Drop<T>> of Serde<Tensor<
 /// dynamic_quantize_linear - Computes the Scale, Zero Point and FP32->8Bit conversion of FP32 Input data. 
 /// scatter_nd - The output of the operation is produced by creating a copy of the input data, and then updating its value to values specified by updates at specific index positions specified by indices. Its output shape is the same as the shape of data
 /// label_encoder - Maps each element in the input tensor to another value.
+/// mel_weight_matrix - Generate a MelWeightMatrix that can be used to re-weight a Tensor containing a linearly sampled frequency spectra (from DFT or STFT) into num_mel_bins frequency information based on the [lower_edge_hertz, upper_edge_hertz] range on the mel scale. 
 trait TensorTrait<T> {
     /// # tensor.new
     ///
@@ -5850,6 +5851,61 @@ trait TensorTrait<T> {
         values: Option<Span<T>>,
         values_tensor: Option<Tensor<T>>
     ) -> Tensor<T>;
+    /// # TensorTrait::mel_weight_matrix
+    /// 
+    /// ```rust
+    ///         fn mel_weight_matrix(num_mel_bins: usize, dft_length: usize, sample_rate: usize, lower_edge_hertz: T, upper_edge_hertz: T) -> Tensor<T>;
+    /// ```
+    ///
+    /// Generate a MelWeightMatrix that can be used to re-weight a Tensor containing a linearly sampled frequency spectra (from DFT or STFT) into num_mel_bins frequency information based on the [lower_edge_hertz, upper_edge_hertz] range on the mel scale. 
+    /// This function defines the mel scale in terms of a frequency in hertz according to the following formula:
+    /// ```
+    /// mel(f) = 2595 * log10(1 + f/700)
+    /// ```
+    /// In the returned matrix, all the triangles (filterbanks) have a peak value of 1.0.
+    /// The returned MelWeightMatrix can be used to right-multiply a spectrogram S of shape [frames, num_spectrogram_bins] of linear scale spectrum values (e.g. STFT magnitudes) to generate a “mel spectrogram” M of shape [frames, num_mel_bins].
+    /// ## Args
+    ///
+    /// * `num_mel_bins `(`usize`) - The number of bands in the mel spectrum.
+    /// * `dft_length `(`usize`) - The size of the original DFT. The size of the original DFT is used to infer the size of the onesided DFT, which is understood to be floor(dft_length/2) + 1, i.e. the spectrogram only contains the nonredundant DFT bins.
+    /// * `sample_rate `(`usize`) - Samples per second of the input signal used to create the spectrogram. Used to figure out the frequencies corresponding to each spectrogram bin, which dictates how they are mapped into the mel scale.
+    /// * `lower_edge_hertz `(T) - Lower bound on the frequencies to be included in the mel spectrum. This corresponds to the lower edge of the lowest triangular band.
+    /// * `upper_edge_hertz  `(T) - The desired top edge of the highest frequency band.
+    ///
+    /// ## Returns
+    /// 
+    /// * A `Tensor<T>` The Mel Weight Matrix. The output has the shape: [floor(dft_length/2) + 1][num_mel_bins].
+    ///
+    /// ## Examples
+    /// 
+    /// ```rust
+    /// use orion::operators::tensor::{FP16x16Tensor, FP16x16TensorAdd};
+    /// use core::array::{ArrayTrait, SpanTrait};
+    /// use orion::operators::tensor::FP16x16TensorPartialEq;
+    /// use orion::utils::{assert_eq, assert_seq_eq};
+    /// use orion::operators::tensor::{TensorTrait, Tensor};
+    /// use orion::numbers::{FixedTrait, FP16x16};
+    ///
+    ///
+    /// fn example() -> Tensor<FP16x16> {
+    ///     return TensorTrait::mel_weight_matrix(8, 16, 8192, FP16x16 { mag: 0, sign: false }, FP16x16 { mag: 268435456, sign: false });
+    /// }
+    /// >>> [
+    ///         [65536, 65536, 0, 0, 0, 0, 0, 0],
+    ///         [0, 0, 65536, 65536, 0, 0, 0, 0],   
+    ///         [0, 0, 0, 0, 65536, 0, 0, 0],
+    ///         [0, 0, 0, 0, 0, 65536, 0, 0],
+    ///         [0, 0, 0, 0, 0, 0, 65536, 0],   
+    ///         [0, 0, 0, 0, 0, 0, 0, 65536],
+    ///         [0, 0, 0, 0, 0, 0, 0, 0],
+    ///         [0, 0, 0, 0, 0, 0, 0, 0],
+    ///         [0, 0, 0, 0, 0, 0, 0, 0],
+    ///     ]
+    /// ```
+    ///
+    fn mel_weight_matrix(
+        num_mel_bins: usize, dft_length: usize, sample_rate: usize, lower_edge_hertz: T, upper_edge_hertz: T
+    ) -> Tensor<T>;
 }
 
 /// Cf: TensorTrait::new docstring

src/operators/tensor/implementations/tensor_bool.cairo

@@ -547,6 +547,12 @@ impl BoolTensor of TensorTrait<bool> {
     ) -> Tensor<bool> {
         panic(array!['not supported!'])
     }
+
+    fn mel_weight_matrix(
+        num_mel_bins: usize, dft_length: usize, sample_rate: usize, lower_edge_hertz: bool, upper_edge_hertz: bool
+    ) -> Tensor<bool>{
+        panic(array!['not supported!'])
+    }
 }
 
 /// Implements partial equal for two `Tensor<bool>` using the `PartialEq` trait.

src/operators/tensor/implementations/tensor_complex64.cairo

@@ -585,6 +585,12 @@ impl Complex64Tensor of TensorTrait<complex64> {
     ) -> Tensor<complex64> {
         panic(array!['not supported!'])
     }
+
+    fn mel_weight_matrix(
+        num_mel_bins: usize, dft_length: usize, sample_rate: usize, lower_edge_hertz: complex64, upper_edge_hertz: complex64
+    ) -> Tensor<complex64>{
+        panic(array!['not supported!'])
+    }
 }
 
 /// Implements addition for `Tensor<complex64>` using the `Add` trait.

src/operators/tensor/implementations/tensor_fp16x16.cairo

@@ -639,6 +639,12 @@ impl FP16x16Tensor of TensorTrait<FP16x16> {
             self, default_list, default_tensor, keys, keys_tensor, values, values_tensor
         )
     }
+
+    fn mel_weight_matrix(
+        num_mel_bins: usize, dft_length: usize, sample_rate: usize, lower_edge_hertz: FP16x16, upper_edge_hertz: FP16x16
+    ) -> Tensor<FP16x16>{
+        math::mel_weight_matrix::mel_weight_matrix(num_mel_bins, dft_length, sample_rate, lower_edge_hertz, upper_edge_hertz)
+    }
 }
 
 /// Implements addition for `Tensor<FP16x16>` using the `Add` trait.

src/operators/tensor/implementations/tensor_fp16x16wide.cairo

@@ -599,6 +599,12 @@ impl FP16x16WTensor of TensorTrait<FP16x16W> {
             self, default_list, default_tensor, keys, keys_tensor, values, values_tensor
         )
     }
+
+    fn mel_weight_matrix(
+        num_mel_bins: usize, dft_length: usize, sample_rate: usize, lower_edge_hertz: FP16x16W, upper_edge_hertz: FP16x16W
+    ) -> Tensor<FP16x16W>{
+        math::mel_weight_matrix::mel_weight_matrix(num_mel_bins, dft_length, sample_rate, lower_edge_hertz, upper_edge_hertz)
+    }
 }
 
 /// Implements addition for `Tensor<FP16x16W>` using the `Add` trait.

src/operators/tensor/implementations/tensor_fp32x32.cairo

@@ -635,6 +635,12 @@ impl FP32x32Tensor of TensorTrait<FP32x32> {
             self, default_list, default_tensor, keys, keys_tensor, values, values_tensor
         )
     }
+
+    fn mel_weight_matrix(
+        num_mel_bins: usize, dft_length: usize, sample_rate: usize, lower_edge_hertz: FP32x32, upper_edge_hertz: FP32x32
+    ) -> Tensor<FP32x32>{
+        math::mel_weight_matrix::mel_weight_matrix(num_mel_bins, dft_length, sample_rate, lower_edge_hertz, upper_edge_hertz)
+    }
 }
 
 /// Implements addition for `Tensor<FP32x32>` using the `Add` trait.

src/operators/tensor/implementations/tensor_fp64x64.cairo

@@ -635,6 +635,12 @@ impl FP64x64Tensor of TensorTrait<FP64x64> {
             self, default_list, default_tensor, keys, keys_tensor, values, values_tensor
         )
     }
+
+    fn mel_weight_matrix(
+        num_mel_bins: usize, dft_length: usize, sample_rate: usize, lower_edge_hertz: FP64x64, upper_edge_hertz: FP64x64
+    ) -> Tensor<FP64x64>{
+        math::mel_weight_matrix::mel_weight_matrix(num_mel_bins, dft_length, sample_rate, lower_edge_hertz, upper_edge_hertz)
+    }
 }
 
 /// Implements addition for `Tensor<FP64x64>` using the `Add` trait.

src/operators/tensor/implementations/tensor_fp8x23.cairo

@@ -633,6 +633,12 @@ impl FP8x23Tensor of TensorTrait<FP8x23> {
             self, default_list, default_tensor, keys, keys_tensor, values, values_tensor
         )
     }
+
+    fn mel_weight_matrix(
+        num_mel_bins: usize, dft_length: usize, sample_rate: usize, lower_edge_hertz: FP8x23, upper_edge_hertz: FP8x23
+    ) -> Tensor<FP8x23>{
+        panic(array!['not supported!'])
+    }
 }
 
 /// Implements addition for `Tensor<FP8x23>` using the `Add` trait.

src/operators/tensor/implementations/tensor_fp8x23wide.cairo

@@ -576,6 +576,12 @@ impl FP8x23WTensor of TensorTrait<FP8x23W> {
             self, default_list, default_tensor, keys, keys_tensor, values, values_tensor
         )
     }
+
+    fn mel_weight_matrix(
+        num_mel_bins: usize, dft_length: usize, sample_rate: usize, lower_edge_hertz: FP8x23W, upper_edge_hertz: FP8x23W
+    ) -> Tensor<FP8x23W>{
+        panic(array!['not supported!'])
+    }
 }
 
 /// Implements addition for `Tensor<FP8x23W>` using the `Add` trait.

src/operators/tensor/implementations/tensor_i32.cairo

@@ -599,6 +599,12 @@ impl I32Tensor of TensorTrait<i32> {
             self, default_list, default_tensor, keys, keys_tensor, values, values_tensor
         )
     }
+
+    fn mel_weight_matrix(
+        num_mel_bins: usize, dft_length: usize, sample_rate: usize, lower_edge_hertz: i32, upper_edge_hertz: i32
+    ) -> Tensor<i32>{
+        panic(array!['not supported!'])
+    }
 }
 
 /// Implements addition for `Tensor<i32>` using the `Add` trait.

src/operators/tensor/implementations/tensor_i8.cairo

@@ -602,6 +602,12 @@ impl I8Tensor of TensorTrait<i8> {
             self, default_list, default_tensor, keys, keys_tensor, values, values_tensor
         )
     }
+
+    fn mel_weight_matrix(
+        num_mel_bins: usize, dft_length: usize, sample_rate: usize, lower_edge_hertz: i8, upper_edge_hertz: i8
+    ) -> Tensor<i8>{
+        panic(array!['not supported!'])
+    }
 }
 
 /// Implements addition for `Tensor<i8>` using the `Add` trait.

src/operators/tensor/implementations/tensor_u32.cairo

@@ -546,6 +546,13 @@ impl U32Tensor of TensorTrait<u32> {
             self, default_list, default_tensor, keys, keys_tensor, values, values_tensor
         )
     }
+
+    fn mel_weight_matrix(
+        num_mel_bins: usize, dft_length: usize, sample_rate: usize, lower_edge_hertz: u32, upper_edge_hertz: u32
+    ) -> Tensor<u32>{
+        // Blocking at the 'log10' operator
+        panic(array!['not supported!'])
+    }
 }
 
 /// Implements addition for `Tensor<u32>` using the `Add` trait.

src/operators/tensor/math.cairo

@@ -68,3 +68,4 @@ mod hann_window;
 mod hamming_window;
 mod blackman_window;
 mod scatter_nd;
+mod mel_weight_matrix;