eden-emu/opus
15
0
Fork 0
mirror of https://github.com/xiph/opus.git synced 2025-05-25 12:49:12 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

wip 8x4 sparseness

Browse source
This commit is contained in:
Jean-Marc Valin 2020-12-19 19:25:59 -05:00
parent 8e405b44e0
commit cc28518699
3 changed files with 13 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

10
dnn/training_tf2/dump_lpcnet.py
View file

@ -66,15 +66,17 @@ def printSparseVector(f, A, name):
A[:,2*N:] = A[:,2*N:] - np.diag(np.diag(A[:,2*N:])) A[:,2*N:] = A[:,2*N:] - np.diag(np.diag(A[:,2*N:]))
printVector(f, diag, name + '_diag') printVector(f, diag, name + '_diag')
idx = np.zeros((0,), dtype='int') idx = np.zeros((0,), dtype='int')
for i in range(3*N//16): for i in range(3*N//8):
pos = idx.shape[0] pos = idx.shape[0]
idx = np.append(idx, -1) idx = np.append(idx, -1)
nb_nonzero = 0 nb_nonzero = 0
for j in range(N): for j in range(N//4):
if np.sum(np.abs(A[j, i*16:(i+1)*16])) > 1e-10: block = A[j*4:(j+1)*4, i*8:(i+1)*8]
if np.sum(np.abs(block)) > 1e-10:
nb_nonzero = nb_nonzero + 1 nb_nonzero = nb_nonzero + 1
idx = np.append(idx, j) idx = np.append(idx, j)
W = np.concatenate([W, A[j, i*16:(i+1)*16]]) vblock = block.transpose((1,0)).reshape((-1,))
W = np.concatenate([W, vblock])
idx[pos] = nb_nonzero idx[pos] = nb_nonzero
printVector(f, W, name) printVector(f, W, name)
#idx = np.tile(np.concatenate([np.array([N]), np.arange(N)]), 3*N//16) #idx = np.tile(np.concatenate([np.array([N]), np.arange(N)]), 3*N//16)

8
dnn/training_tf2/lpcnet.py
View file

@ -74,12 +74,14 @@ class Sparsify(Callback):
A = p[:, k*N:(k+1)*N] A = p[:, k*N:(k+1)*N]
A = A - np.diag(np.diag(A)) A = A - np.diag(np.diag(A))
#A = np.transpose(A, (1, 0)) #A = np.transpose(A, (1, 0))
L=np.reshape(A, (N, N//16, 16)) L=np.reshape(A, (N//4, 4, N//8, 8))
S=np.sum(L*L, axis=-1) S=np.sum(L*L, axis=-1)
S=np.sum(S, axis=1)
SS=np.sort(np.reshape(S, (-1,))) SS=np.sort(np.reshape(S, (-1,)))
thresh = SS[round(N*N//16*(1-density))] thresh = SS[round(N*N//32*(1-density))]
mask = (S>=thresh).astype('float32'); mask = (S>=thresh).astype('float32');
mask = np.repeat(mask, 16, axis=1) mask = np.repeat(mask, 4, axis=0)
mask = np.repeat(mask, 8, axis=1)
mask = np.minimum(1, mask + np.diag(np.ones((N,)))) mask = np.minimum(1, mask + np.diag(np.ones((N,))))
#mask = np.transpose(mask, (1, 0)) #mask = np.transpose(mask, (1, 0))
p[:, k*N:(k+1)*N] = p[:, k*N:(k+1)*N]*mask p[:, k*N:(k+1)*N] = p[:, k*N:(k+1)*N]*mask

4
dnn/training_tf2/train_lpcnet.py
View file

@ -102,7 +102,7 @@ del pred
del in_exc del in_exc
# dump models to disk as we go # dump models to disk as we go
checkpoint = ModelCheckpoint('lpcnet32c_384_10_G16_{epoch:02d}.h5') checkpoint = ModelCheckpoint('lpcnet32v_384_10_G16_{epoch:02d}.h5')
#Set this to True to adapt an existing model (e.g. on new data) #Set this to True to adapt an existing model (e.g. on new data)
adaptation = False adaptation = False
@ -120,5 +120,5 @@ else:
decay = 5e-5 decay = 5e-5
model.compile(optimizer=Adam(lr, decay=decay, beta_2=0.99), loss='sparse_categorical_crossentropy') model.compile(optimizer=Adam(lr, decay=decay, beta_2=0.99), loss='sparse_categorical_crossentropy')
model.save_weights('lpcnet32c_384_10_G16_00.h5'); model.save_weights('lpcnet32v_384_10_G16_00.h5');
model.fit([in_data, features, periods], out_exc, batch_size=batch_size, epochs=nb_epochs, validation_split=0.0, callbacks=[checkpoint, sparsify]) model.fit([in_data, features, periods], out_exc, batch_size=batch_size, epochs=nb_epochs, validation_split=0.0, callbacks=[checkpoint, sparsify])