# jobicari/noisy-speech-asr-benchmarking-pipeline architecture

> Architecture diagram of the GitHub repository jobicari/noisy-speech-asr-benchmarking-pipeline, generated by GitDiagram from its file tree, README and a sample of source files (last updated 2026-11-16). It is an AI-made overview: check details against the source.

- Interactive diagram: https://gitdiagram.com/jobicari/noisy-speech-asr-benchmarking-pipeline
- Repository: https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline

## Overview

Noisy-Speech ASR Benchmark measures recognition accuracy across clean and synthetically damaged LibriSpeech audio. The main run loads speech and an impulse response, prepares seeded corruption variants and audio-health metrics in CPU workers, batches them through configured ASR models, then evaluates transcripts and checkpoints results. Configuration and run metadata support reproducibility; analysis and visualization present benchmark findings. The README documents cloud execution and result upload, but sampled source does not show that orchestration, so those links are omitted.

Main entry points: [`app.py`](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/app.py) (Run entry point), [`src/asr_pipeline_benchmark/models.py`](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/models.py) (Transcriber engines), [`src/asr_pipeline_benchmark/prep.py`](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/prep.py) (Worker preparation), [`src/asr_pipeline_benchmark/storage.py`](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/storage.py) (Checkpoint storage)

## Diagram (Mermaid)

```mermaid
flowchart TD

subgraph group_run["Benchmark run"]
  node_entry["Run entry point<br/>[app.py]"]
  node_settings["Run configuration<br/>[configuration.py]"]
  node_speech["Speech dataset<br/>[data.py]"]
  node_utility["Run metadata and batching<br/>[utility.py]"]
end

subgraph group_audio["Audio preparation"]
  node_impulse["Impulse dataset<br/>[data.py]"]
  node_audio_prep["Corruptions and cleanup<br/>[preprocessing.py]"]
  node_worker_prep["Worker preparation<br/>[prep.py]"]
  node_health["Audio health metrics<br/>[audio_analysis.py]"]
end

subgraph group_asr["Recognition and metrics"]
  node_models["Transcriber engines<br/>[models.py]"]
  node_evaluation["Error-rate evaluation<br/>[evaluation.py]"]
end

subgraph group_results["Results and analysis"]
  node_storage_api["Checkpoint storage<br/>[storage.py]"]
  node_analysis["Benchmark analysis"]
  node_plots["Feature visualization<br/>[visualization.py]"]
end

node_engineer(("Benchmark operator"))
node_dataset_service["Hugging Face datasets"]
node_asr_provider["ASR model runtimes"]
node_local_store[("Local files")]
node_gcs[("Google Cloud Storage")]

node_engineer -->|"starts run"| node_entry
node_entry -->|"reads settings"| node_settings
node_entry -->|"loads speech"| node_speech
node_entry -->|"loads impulse response"| node_audio_prep
node_speech -->|"streams dataset"| node_dataset_service
node_impulse -->|"loads dataset"| node_dataset_service
node_entry -->|"initializes workers"| node_worker_prep
node_entry -->|"runs checkpoint loop"| node_models
node_entry -->|"selects backend"| node_storage_api
node_entry -->|"records run"| node_utility
node_worker_prep -->|"applies corruptions"| node_audio_prep
node_worker_prep -->|"analyzes audio"| node_health
node_worker_prep -->|"seeds and times"| node_utility
node_models -->|"dispatches preparation"| node_worker_prep
node_models -->|"transcribes batches"| node_asr_provider
node_models -->|"checkpoints results"| node_storage_api
node_models -->|"batches samples"| node_utility
node_evaluation -->|"uses transcripts"| node_models
node_storage_api -->|"reads and writes"| node_local_store
node_storage_api -->|"reads and writes"| node_gcs
node_analysis -.->|"analyzes run results"| node_storage_api
node_plots -.->|"plots features"| node_analysis

click node_entry "https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/app.py"
click node_settings "https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/configuration.py"
click node_speech "https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/data.py"
click node_impulse "https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/data.py"
click node_audio_prep "https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/preprocessing.py"
click node_worker_prep "https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/prep.py"
click node_health "https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/audio_analysis.py"
click node_models "https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/models.py"
click node_evaluation "https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/evaluation.py"
click node_utility "https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/utility.py"
click node_storage_api "https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/storage.py"
click node_analysis "https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/tree/main/analysis/v2"
click node_plots "https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/visualization.py"

classDef toneNeutral fill:#f8fafc,stroke:#334155,stroke-width:1.5px,color:#0f172a
classDef toneBlue fill:#dbeafe,stroke:#2563eb,stroke-width:1.5px,color:#172554
classDef toneAmber fill:#fef3c7,stroke:#d97706,stroke-width:1.5px,color:#78350f
classDef toneMint fill:#dcfce7,stroke:#16a34a,stroke-width:1.5px,color:#14532d
classDef toneRose fill:#ffe4e6,stroke:#e11d48,stroke-width:1.5px,color:#881337
classDef toneIndigo fill:#e0e7ff,stroke:#4f46e5,stroke-width:1.5px,color:#312e81
classDef toneTeal fill:#ccfbf1,stroke:#0f766e,stroke-width:1.5px,color:#134e4a
class node_entry,node_settings,node_speech,node_utility toneBlue
class node_impulse,node_audio_prep,node_worker_prep,node_health,node_local_store,node_gcs toneAmber
class node_models,node_evaluation toneMint
class node_storage_api,node_analysis,node_plots toneRose
class node_engineer,node_dataset_service,node_asr_provider toneIndigo
```

## Components

### Benchmark run

- [Run entry point](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/app.py): `app.py`
- [Run configuration](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/configuration.py): `src/asr_pipeline_benchmark/configuration.py`
- [Speech dataset](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/data.py): `src/asr_pipeline_benchmark/data.py`
- [Run metadata and batching](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/utility.py): `src/asr_pipeline_benchmark/utility.py`

### Audio preparation

- [Impulse dataset](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/data.py): `src/asr_pipeline_benchmark/data.py`
- [Corruptions and cleanup](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/preprocessing.py): `src/asr_pipeline_benchmark/preprocessing.py`
- [Worker preparation](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/prep.py): `src/asr_pipeline_benchmark/prep.py`
- [Audio health metrics](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/audio_analysis.py): `src/asr_pipeline_benchmark/audio_analysis.py`

### Recognition and metrics

- [Transcriber engines](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/models.py): `src/asr_pipeline_benchmark/models.py`
- [Error-rate evaluation](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/evaluation.py): `src/asr_pipeline_benchmark/evaluation.py`

### Results and analysis

- [Checkpoint storage](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/storage.py): `src/asr_pipeline_benchmark/storage.py`
- [Benchmark analysis](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/tree/main/analysis/v2): `analysis/v2`
- [Feature visualization](https://github.com/jobicari/noisy-speech-asr-benchmarking-pipeline/blob/main/src/asr_pipeline_benchmark/visualization.py): `src/asr_pipeline_benchmark/visualization.py`

### Other components

- **Benchmark operator**
- **Hugging Face datasets**
- **ASR model runtimes**
- **Local files**
- **Google Cloud Storage**

## Connections

- Benchmark operator → Run entry point: starts run (evidence: `README.md`)
- Run entry point → Run configuration: reads settings (evidence: `app.py`)
- Run entry point → Speech dataset: loads speech (evidence: `app.py`)
- Run entry point → Corruptions and cleanup: loads impulse response (evidence: `app.py`)
- Speech dataset → Hugging Face datasets: streams dataset (evidence: `src/asr_pipeline_benchmark/data.py`)
- Impulse dataset → Hugging Face datasets: loads dataset (evidence: `src/asr_pipeline_benchmark/data.py`)
- Run entry point → Worker preparation: initializes workers (evidence: `app.py`)
- Run entry point → Transcriber engines: runs checkpoint loop (evidence: `app.py`)
- Run entry point → Checkpoint storage: selects backend (evidence: `app.py`)
- Run entry point → Run metadata and batching: records run (evidence: `app.py`)
- Worker preparation → Corruptions and cleanup: applies corruptions (evidence: `src/asr_pipeline_benchmark/prep.py`)
- Worker preparation → Audio health metrics: analyzes audio (evidence: `src/asr_pipeline_benchmark/prep.py`)
- Worker preparation → Run metadata and batching: seeds and times (evidence: `src/asr_pipeline_benchmark/prep.py`)
- Transcriber engines → Worker preparation: dispatches preparation (evidence: `src/asr_pipeline_benchmark/models.py`)
- Transcriber engines → ASR model runtimes: transcribes batches (evidence: `README.md`)
- Transcriber engines → Checkpoint storage: checkpoints results (evidence: `src/asr_pipeline_benchmark/models.py`)
- Transcriber engines → Run metadata and batching: batches samples (evidence: `src/asr_pipeline_benchmark/models.py`)
- Error-rate evaluation → Transcriber engines: uses transcripts (evidence: `src/asr_pipeline_benchmark/evaluation.py`)
- Checkpoint storage → Local files: reads and writes (evidence: `src/asr_pipeline_benchmark/storage.py`)
- Checkpoint storage → Google Cloud Storage: reads and writes (evidence: `src/asr_pipeline_benchmark/storage.py`)
- Benchmark analysis → Checkpoint storage: analyzes run results
- Feature visualization → Benchmark analysis: plots features
