| | |
|---|
| | | |
|---|
| | | |
|---|
| | | def calculate_embedding(spk, spk2utts, utt2xvec, embedding_dim, average_emb_num): |
|---|
| | | # process for empty speaker |
|---|
| | | # process for dummy speaker |
|---|
| | | if spk == "None": |
|---|
| | | return np.zeros((embedding_dim, ), dtype=np.float32) |
|---|
| | | return np.zeros((1, embedding_dim), dtype=np.float32) |
|---|
| | | |
|---|
| | | # calculate averaged speaker embeddings |
|---|
| | | utt_list = spk2utts[spk] |
|---|
| | |
|---|
| | | xvec_list = [kaldiio.load_mat(utt2xvec[utt]) for utt in utt_list] |
|---|
| | | else: |
|---|
| | | xvec_list = [kaldiio.load_mat(utt2xvec[utt]) for utt in random.sample(utt_list, average_emb_num)] |
|---|
| | | xvec = np.mean(np.hstack(xvec_list), axis=0) |
|---|
| | | xvec = np.mean(np.concatenate(xvec_list, axis=0), axis=0) |
|---|
| | | |
|---|
| | | return xvec |
|---|
| | | |
|---|
