1     Functions

2     Reference Architecture

3     I/O Data of AI Workflow

4     Functions of AI Modules

5     I/O Data of AI Modules

6     AIWs, AIMs, and JSON Metadata

1      Functions

The goal of this use case is to restore a Damaged Segment of an Audio Segment containing only speech from a single speaker. The damage may affect the entire segment, or only part of it.

Restoration will not involve filtering or signal processing. Instead, replacements for the damaged vocal elements will be synthesised using a speech model. The latter is a component or set of components, normally including one or more neural networks, which accepts text and possibly other specifications, and delivers audible speech in a specified format – here, the speech of the required replacement or replacements. If the damage affects the entire segment, an entirely new segment is synthesized; if only parts are affected, corresponding segments will be synth­esized individually to enable later integration into the undamaged parts of the Damaged Segment, with reference to appropriate Time Labels.

The speech segments necessary for the creation of the speech model can be flexibly resourced from undamaged parts of the input segment or from other recording sources that are consistent with the original segment’s acoustic environment.

Restoration is carried out by synthesising replacements for the damaged vocal elements as follows:

The Speech Segments for Modelling – Audio Segments necessary for the creation of the Neural Network Speech Model – may be obtained from any undamaged parts of the input speech segment; however, other Audio Segments consistent with the original segment’s sound environment can also be used.

2      Reference Architecture

The Reference Model of the Speech Restoration System is given by Figure 1.

Figure 1 – Speech Restoration System (SRS) Reference Model

In the SRS use case, the entire Damaged Segment can be replaced by a synthesised segment, or parts within it can be synthesized to enable integration of the replaced segments.

The sequence of events in this Use Case is as follows:

  1. Speech Model Creation receives Audio Segments for Modelling, a set of recordings composing a corpus that will be used to train a Neural Network Speech Model in Speech Model Creation.
  2. That Neural Network Speech Model is passed to the Speech Synthesis for Restoration AIM, which also receives a Text List as input. Each element of Text List is a string specifying the text of a damaged section of Damaged Segment (or of Damaged Segment as a whole). Speech Synthesis for Restoration produces synthetic replacements for each damaged section (or for Damaged Segment as a whole) and passes the replacement(s) to Speech Restoration Assembly.
  3. Speech Restoration Assembly receives as input the entire Damaged Segment, plus Damaged List, a list indicating the locations of any damaged sections within Damaged Segment. The list will be null if Damaged Segment in its entirety was replaced.
  4. Speech Restoration Assembly produces as output Restored Segment, in which any repaired sections have been replaced by synthetic sections, or in which the entire Damaged Segment has been replaced.

3      I/O Data of AI Workflow

Table 1 gives the input and output data of Speech Restoration System.

Table 1 – I/O data of Audio Recording Preservation

Input Comments
Speech Segments for Modelling A set of Audio Files containing speech segments used to train the Neural Network Speech Model.
Text List List of texts to be converted into speech by the Speech Synthesis for Restoration AIM.
Damaged List A list of strings of Texts corresponding to the Damaged Segments (if any) requiring replacement with synthetic segments.
Damaged Segment An Audio Segment containing only speech (and not containing music or other sounds) which is either damaged in its entirety or contains one or more Damaged Sections specified in the Damaged List.
Output Comments
Restored Speech Segment An Audio Segment in which the entire segment has been replaced by a synthetic speech segment, or in which each Damaged Segment has been replaced by a synthetic speech segment.

4      Functions of AI Modules

The AIMs required by the Speech Restoration System Use Case are described in Table 2/

Table 2 – AI Modules of Speech Recording System

AIM Function
Speech Model Creation 1.     Receives in separate files the Audio Segments for Modelling, adequate for model creation.
2.     Creates the current Neural Network Speech Model.
3.     Sends that Neural Network Speech Model to the Speech Synthesis for Restoration.
Speech Synthesis for Restoration 1.     Receives the current Neural Network Speech Model.
2.     Receives Damaged List as a data structure:
a.     Containing one element if Damaged Segment is damaged throughout or
b.     Representing a list in which each element specifies via Time Lab­els the start and end of a damaged section within Damaged Seg­ment.
3.     Synthesizes each Damaged Section in Damaged List.
4.     Sends the newly synthesised segments to the Speech Restoration Assembly as an ordered list.
Speech Restoration Assembly 1.     Receives the Damaged Segment.
2.     Receives the ordered list of synthetic segments.
3.     Receives Damaged List Time Labels, indicating where the synthesized segments should be inserted in left-to-right order. In case Damaged Segment as a whole was damaged, the list contains one entry.
4.     Assembles the final version of the Restored Segment.

5      I/O Data of AI Modules

Table 3 – CAE-SRS AIMs and their I/O Data

AIM Input Data Output Data
Speech Model Creation Audio Segments for Modelling Neural Network Speech Model
Speech Synthesis for Restoration Text List
Neural Network Speech Model
Synthesised Speech
Speech Restoration Assembly Damaged Segments
Damaged List
Restored Segment

6      AIW, AIMs, and JSON Metadata

Table 4 – AIMs and JSON Metadata

CAE-SRS Speech Restoration System File
CAE-SMC Speech Model Creation File
CAE-SSR Speech Synthesis for Restoration File
CAE-SRA Speech Restoration Assembly File