From 0eb8911a29469f1c6c17c06e46917f2947e27cde Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Herv=C3=A9=20BREDIN?= Date: Tue, 8 Apr 2025 22:31:24 +0200 Subject: [PATCH] feat: add SSeRiouSS.wav2vec_only property --- .../audio/models/segmentation/SSeRiouSS.py | 18 +++++++++++++++++- 1 file changed, 17 insertions(+), 1 deletion(-) diff --git a/src/pyannote/audio/models/segmentation/SSeRiouSS.py b/src/pyannote/audio/models/segmentation/SSeRiouSS.py index bd69c568..81bc97e5 100644 --- a/src/pyannote/audio/models/segmentation/SSeRiouSS.py +++ b/src/pyannote/audio/models/segmentation/SSeRiouSS.py @@ -286,6 +286,16 @@ class SSeRiouSS(Model): ) return receptive_field_center + + @property + def wav2vec_only(self) -> bool: + """Whether to only extract wav2vec features""" + return getattr(self, "_wav2vec_only", False) + + @wav2vec_only.setter + def wav2vec_only(self, value: bool): + self._wav2vec_only = value + def forward(self, waveforms: torch.Tensor) -> torch.Tensor: """Pass forward @@ -295,7 +305,9 @@ class SSeRiouSS(Model): Returns ------- - scores : (batch, frame, classes) + output : (batch, frame, classes) or (batch, frame, wav2vec_dimension) + If `self.wav2vec_only` is True, returns wav2vec features (after weighted average) + If `self.wav2vec_only` is False, returns segmentation scores """ num_layers = ( @@ -313,6 +325,10 @@ class SSeRiouSS(Model): else: outputs = outputs[-1] + if self.wav2vec_only: + return outputs + + if self.hparams.lstm["monolithic"]: outputs, _ = self.lstm(outputs) else: