SRMR: update note and change default parameters (#1872)

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Jirka Borovec <6035284+Borda@users.noreply.github.com>

CHANGELOG.md

@@ -11,7 +11,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
-- Added speech-to-reverberation modulation energy ratio (SRMR) metric ([#1792](https://github.com/Lightning-AI/torchmetrics/pull/1792))
+- Added speech-to-reverberation modulation energy ratio (SRMR) metric ([#1792](https://github.com/Lightning-AI/torchmetrics/pull/1792), [#1872](https://github.com/Lightning-AI/torchmetrics/pull/1872))
 
 
 - Added new global arg `compute_with_cache` to control caching behaviour after `compute` method ([#1754](https://github.com/Lightning-AI/torchmetrics/pull/1754))

src/torchmetrics/audio/srmr.py

@@ -53,6 +53,12 @@ class SpeechReverberationModulationEnergyRatio(Metric):
         Either install as ``pip install torchmetrics[audio]`` or ``pip install torchaudio``
         and ``pip install git+https://github.com/detly/gammatone``.
 
+    .. note::
+        This implementation is experimental, and might not be consistent with the matlab
+        implementation `SRMRToolbox`_, especially the fast implementation.
+        The slow versions, a) fast=False, norm=False, max_cf=128, b) fast=False, norm=True, max_cf=30, have
+        a relatively small inconsistence.
+
     Args:
         fs: the sampling rate
         n_cochlear_filters: Number of filters in the acoustic filterbank
@@ -93,7 +99,7 @@ def __init__(
         n_cochlear_filters: int = 23,
         low_freq: float = 125,
         min_cf: float = 4,
-        max_cf: Optional[float] = 128,
+        max_cf: Optional[float] = None,
         norm: bool = False,
         fast: bool = False,
         **kwargs: Any,

src/torchmetrics/functional/audio/srmr.py

@@ -183,7 +183,7 @@ def speech_reverberation_modulation_energy_ratio(
     n_cochlear_filters: int = 23,
     low_freq: float = 125,
     min_cf: float = 4,
-    max_cf: Optional[float] = 128,
+    max_cf: Optional[float] = None,
     norm: bool = False,
     fast: bool = False,
 ) -> Tensor:
@@ -210,6 +210,12 @@ def speech_reverberation_modulation_energy_ratio(
         Either install as ``pip install torchmetrics[audio]`` or ``pip install torchaudio``
         and ``pip install git+https://github.com/detly/gammatone``.
 
+    .. note::
+        This implementation is experimental, and might not be consistent with the matlab
+        implementation `SRMRToolbox`_, especially the fast implementation.
+        The slow versions, a) fast=False, norm=False, max_cf=128, b) fast=False, norm=True, max_cf=30, have
+        a relatively small inconsistence.
+
     Returns:
         Tensor: srmr value, shape ``(...)``
 

tests/unittests/audio/test_srmr.py

@@ -39,7 +39,7 @@ def _ref_metric_batch(preds: Tensor, target: Tensor, fs: int, fast: bool, norm:
     preds = preds.detach().cpu().numpy()
     score = []
     for b in range(preds.shape[0]):
-        val, _ = srmrpy_srmr(preds[b, ...], fs=fs, fast=fast, norm=norm)
+        val, _ = srmrpy_srmr(preds[b, ...], fs=fs, fast=fast, norm=norm, max_cf=128 if not norm else 30)
         score.append(val)
     score = torch.tensor(score)
     return score.reshape(*shape[:-1])