From 16c21c46cb39565d8615cf0708d6710ace3eb816 Mon Sep 17 00:00:00 2001 From: Albert Zeyer Date: Fri, 7 Feb 2025 11:54:50 +0100 Subject: [PATCH] doc --- .../standalone/model_2024_ctc_spm10k.py | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-) diff --git a/users/zeyer/experiments/exp2024_04_23_baselines/standalone/model_2024_ctc_spm10k.py b/users/zeyer/experiments/exp2024_04_23_baselines/standalone/model_2024_ctc_spm10k.py index ec0e51a95..40544ad2f 100644 --- a/users/zeyer/experiments/exp2024_04_23_baselines/standalone/model_2024_ctc_spm10k.py +++ b/users/zeyer/experiments/exp2024_04_23_baselines/standalone/model_2024_ctc_spm10k.py @@ -315,7 +315,15 @@ def __call__( Use :func:`log_probs_wb_from_logits` to get log probs (might be just log_softmax, but there are some other cases). - :return: logits, enc, enc_spatial_dim + :param source: shape {..., in_spatial_dim}, for example {batch_dim, in_spatial_dim}, audio samples, 16khz + :param in_spatial_dim: input spatial dim + :param collected_outputs: if provided, will write intermediate encoder outputs into it + :return: logits, enc, enc_spatial_dim. + logits shape: {..., enc_spatial_dim, self.wb_target_dim}. + enc shape: {..., enc_spatial_dim, self.encoder.out_dim}. + enc_spatial_dim: output spatial dim. + Use :func:`log_probs_wb_from_logits` on the logits to get log probs + (which is just :func:`rf.log_softmax` in the standard case). """ # log mel filterbank features source, in_spatial_dim = rf.audio.log_mel_filterbank_from_raw(