athena.solver

high-level abstraction of different stages in speech processing

Module Contents

Classes

BaseSolver	Base Training Solver.
HorovodSolver	A multi-processer solver based on Horovod
DecoderSolver	ASR DecoderSolver
SynthesisSolver	SynthesisSolver (TTS Solver)
VadSolver	VadSolver
AVSolver	Base Solver.
AVHorovodSolver	A multi-processer solver based on Horovod
AVDecoderSolver	DecoderSolver

class athena.solver.BaseSolver(model, optimizer, sample_signature, eval_sample_signature=None, config=None, **kwargs)

Bases: tensorflow.keras.Model

Base Training Solver.

default_config

static initialize_devices(solver_gpus=None)

initialize hvd devices, should be called firstly

static clip_by_norm(grads, norm)

clip norm using tf.clip_by_norm

train_step(samples)

train the model 1 step

train(trainset, devset, checkpointer, pbar, epoch, total_batches=-1)

Update the model in 1 epoch

save_checkpointer(checkpointer, devset, epoch)

evaluate_step(samples)

evaluate the model 1 step

evaluate(dataset, epoch)

evaluate the model

class athena.solver.HorovodSolver(model, optimizer, sample_signature, eval_sample_signature=None, config=None, **kwargs)

Bases: BaseSolver

A multi-processer solver based on Horovod

static initialize_devices(solver_gpus=None)

initialize hvd devices, should be called firstly

For examples, if you have two machines and each of them contains 4 gpus: 1. run with command horovodrun -np 6 -H ip1:2,ip2:4 and set solver_gpus to be [0,3,0,1,2,3],

then the first gpu and the last gpu on machine1 and all gpus on machine2 will be used.

2. run with command horovodrun -np 6 -H ip1:2,ip2:4 and set solver_gpus to be [], then the first 2 gpus on machine1 and all gpus on machine2 will be used.

Parameters

solver_gpus ([list]) – a list to specify gpus being used.

Raises

ValueError – If the list of solver gpus is not empty, its size should not be smaller than that of horovod configuration.

train_step(samples)

train the model 1 step

train(trainset, devset, checkpointer, pbar, epoch, total_batches=-1)

Update the model in 1 epoch

evaluate(dataset, epoch=0)

evaluate the model

class athena.solver.DecoderSolver(model, data_descriptions=None, config=None)

Bases: BaseSolver

ASR DecoderSolver

default_config

inference(dataset_builder, rank_size=1, conf=None)

decode the model

inference_saved_model(dataset_builder, rank_size=1, conf=None)

decode the model

class athena.solver.SynthesisSolver(model, optimizer=None, sample_signature=None, eval_sample_signature=None, config=None, **kwargs)

Bases: BaseSolver

SynthesisSolver (TTS Solver)

default_config

inference(dataset_builder, rank_size=1, conf=None)

synthesize using vocoder on dataset

inference_saved_model(dataset_builder, rank_size=1, conf=None)

synthesize using vocoder on dataset

class athena.solver.VadSolver(model, optimizer=None, sample_signature=None, eval_sample_signature=None, data_descriptions=None, config=None)

Bases: BaseSolver

VadSolver

default_config

inference(dataset, rank_size=1, conf=None)

decode the model

class athena.solver.AVSolver(model, optimizer, sample_signature, eval_sample_signature=None, config=None, **kwargs)

Bases: tensorflow.keras.Model

Base Solver.

default_config

static initialize_devices(solver_gpus=None)

initialize hvd devices, should be called firstly

static clip_by_norm(grads, norm)

clip norm using tf.clip_by_norm

train_step(samples)

train the model 1 step

train(trainset, devset, checkpointer, pbar, epoch, total_batches=-1)

Update the model in 1 epoch

evaluate_step(samples)

evaluate the model 1 step

evaluate(dataset, epoch)

evaluate the model

class athena.solver.AVHorovodSolver(model, optimizer, sample_signature, eval_sample_signature=None, config=None, **kwargs)

Bases: AVSolver

A multi-processer solver based on Horovod

static initialize_devices(solver_gpus=None)

initialize hvd devices, should be called firstly

For examples, if you have two machines and each of them contains 4 gpus: 1. run with command horovodrun -np 6 -H ip1:2,ip2:4 and set solver_gpus to be [0,3,0,1,2,3],

then the first gpu and the last gpu on machine1 and all gpus on machine2 will be used.

2. run with command horovodrun -np 6 -H ip1:2,ip2:4 and set solver_gpus to be [], then the first 2 gpus on machine1 and all gpus on machine2 will be used.

Parameters

solver_gpus ([list]) – a list to specify gpus being used.

Raises

ValueError – If the list of solver gpus is not empty, its size should not be smaller than that of horovod configuration.

train_step(samples)

train the model 1 step

train(trainset, devset, checkpointer, pbar, epoch, total_batches=-1)

Update the model in 1 epoch

evaluate(dataset, epoch=0)

evaluate the model

class athena.solver.AVDecoderSolver(model, data_descriptions=None, config=None)

Bases: AVSolver

DecoderSolver

default_config

inference(dataset_builder, rank_size=1, conf=None)

decode the model

inference_freeze(dataset_builder, rank_size=1, conf=None)

decode the model

inference_argmax(dataset_builder, rank_size=1, conf=None)

decode the model