lstmutils

Documentation

Overview

Package lstmutils contains some useful functions for processing trained LSTM models in the particular format used by aural2.

Index

• func LoadGraph(graph *tf.Graph, sess *tf.Session, scopeName string) (input, output tf.Output, statePlaceholders, stateFetches []tf.Output, ...)
• func MakeRenderLSTMstate(oSession *tftrain.OnlineSess) (renderLSTMstate func(*tf.Tensor) ([]byte, error), err error)
• func MakeSeqInference(savedModel tf.SavedModel) (seqInference func(*tf.Tensor) (*tf.Tensor, error), err error)
• func MakeStepInference(oSession tftrain.OnlineSess) (stepInference func(*tf.Tensor) ([]float32, error), err error)

Functions

func LoadGraph

func LoadGraph(graph *tf.Graph, sess *tf.Session, scopeName string) (input, output tf.Output, statePlaceholders, stateFetches []tf.Output, stateFeeds map[tf.Output]*tf.Tensor, err error)

LoadGraph loads the TF graph from the bytes of an LSTM graphdef, and returns a

func MakeRenderLSTMstate

func MakeRenderLSTMstate(oSession *tftrain.OnlineSess) (renderLSTMstate func(*tf.Tensor) ([]byte, error), err error)

func MakeSeqInference

func MakeSeqInference(savedModel tf.SavedModel) (seqInference func(*tf.Tensor) (*tf.Tensor, error), err error)

MakeSeqInference returns a function that takes an output of a slice of mfccs of one clip, and returns a [][]float32 labels.

func MakeStepInference

func MakeStepInference(oSession tftrain.OnlineSess) (stepInference func(*tf.Tensor) ([]float32, error), err error)

MakeStepInference returns a function that takes a tensor of one mfccs, and returns a []float32 labels.