Skip to content

mateusoliveiradev1/frescari / .agents/skills/audio-transcriber

audio-transcriber

Verified

Transform audio recordings into professional Markdown documentation with intelligent summaries using LLM integration

FileKindSize
CHANGELOG.mdmarkdown5.3 KB
README.mdmarkdown9.3 KB
SKILL.mdskill md14.9 KB
examples/basic-transcription.shscript6.3 KB
references/tools-comparison.mdmarkdown8.2 KB
scripts/install-requirements.shscript6.7 KB
scripts/transcribe.pyscript17.3 KB

references/tools-comparison.md

View on GitHub
# Transcription Tools Comparison

Comprehensive comparison of audio transcription engines supported by the audio-transcriber skill.

## Overview

| Tool | Type | Speed | Quality | Cost | Privacy | Offline | Languages |
|------|------|-------|---------|------|---------|---------|-----------|
| **Faster-Whisper** | Open-source | ⚡⚡⚡⚡⚡ | ⭐⭐⭐⭐⭐ | Free | 100% | ✅ | 99 |
| **Whisper** | Open-source | ⚡⚡⚡ | ⭐⭐⭐⭐⭐ | Free | 100% | ✅ | 99 |
| Google Speech-to-Text | Commercial API | ⚡⚡⚡⚡ | ⭐⭐⭐⭐⭐ | $0.006/15s | Partial | ❌ | 125+ |
| Azure Speech | Commercial API | ⚡⚡⚡⚡ | ⭐⭐⭐⭐ | $1/hour | Partial | ❌ | 100+ |
| AssemblyAI | Commercial API | ⚡⚡⚡⚡ | ⭐⭐⭐⭐⭐ | $0.00025/s | Partial | ❌ | 99 |

---

## Faster-Whisper (Recommended)

### Pros
✅ **4-5x faster** than original Whisper  
✅ **Same quality** as original Whisper  
✅ **Lower memory usage** (50-60% less RAM)  
✅ **Free and open-source**  
✅ **100% offline** (privacy guaranteed)  
✅ **Easy installation** (`pip install faster-whisper`)  
✅ **Drop-in replacement** for Whisper

### Cons
❌ Requires Python 3.8+  
❌ Initial model download (~100MB-1.5GB)  
❌ GPU optional but speeds up significantly

### Installation

```bash
pip install faster-whisper
```

### Usage Example

```python
from faster_whisper import WhisperModel

# Load model (auto-downloads on first run)
model = WhisperModel("base", device="cpu", compute_type="int8")

# Transcribe
segments, info = model.transcribe("audio.mp3", language="pt")

# Print results
for segment in segments:
    print(f"[{segment.start:.2f}s -> {segment.end:.2f}s] {segment.text}")
```

### Model Sizes

| Model | Size | RAM | Speed (CPU) | Quality |
|-------|------|-----|-------------|---------|
| `tiny` | 39 MB | ~1 GB | Very fast (~10x realtime) | Basic |
| `base` | 74 MB | ~1 GB | Fast (~7x realtime) | Good |
| `small` | 244 MB | ~2 GB | Moderate (~4x realtime) | Very good |
| `medium` | 769 MB | ~5 GB | Slow (~2x realtime) | Excellent |
| `large` | 1550 MB | ~10 GB | Very slow (~1x realtime) | Best |

**Recommendation:** `small` or `medium` for production use.

---

## Whisper (Original)

### Pros
✅ **Official OpenAI model**  
✅ **Excellent quality**  
✅ **Free and open-source**  
✅ **100% offline**  
✅ **Well-documented**  
✅ **Large community**

### Cons
❌ **Slower** than Faster-Whisper (4-5x)  
❌ **Higher memory usage**  
❌ Requires PyTorch (large dependency)  
❌ GPU highly recommended for larger models

### Installation

```bash
pip install openai-whisper
```

### Usage Example

```python
import whisper

# Load model
model = whisper.load_model("base")

# Transcribe
result = model.transcribe("audio.mp3", language="pt")

# Print results
print(result["text"])
```

### When to Use Whisper vs. Faster-Whisper

**Use Faster-Whisper if:**
- Speed is important
- Limited RAM available
- Processing many files

**Use Original Whisper if:**
- Faster-Whisper installation issues
- Need exact OpenAI implementation
- Already have Whisper in project dependencies

---

## Google Cloud Speech-to-Text

### Pros
✅ **Very accurate** (industry-leading)  
✅ **Fast processing** (cloud infrastructure)  
✅ **125+ languages**  
✅ **Word-level timestamps**  
✅ **Punctuation & capitalization**  
✅ **Speaker diarization** (premium)

### Cons
❌ **Requires internet** (cloud-only)  
❌ **Costs money** (after free tier)  
❌ **Privacy concerns** (audio uploaded to Google)  
❌ Requires GCP account setup  
❌ Complex authentication

### Pricing

- **Free tier:** 60 minutes/month
- **Standard:** $0.006 per 15 seconds ($1.44/hour)
- **Premium:** $0.009 per 15 seconds (with diarization)

### Installation

```bash
pip install google-cloud-speech
```

### Setup

1. Create GCP project
2. Enable Speech-to-Text API
3. Create service account & download JSON key
4. Set environment variable:
   ```bash
   export GOOGLE_APPLICATION_CREDENTIALS="path/to/key.json"
   ```

### Usage Example

```python
from google.cloud import speech

client = speech.SpeechClient()

with open("audio.wav", "rb") as audio_file:
    content = audio_file.read()

audio = speech.RecognitionAudio(content=content)
config = speech.RecognitionConfig(
    encoding=speech.RecognitionConfig.AudioEncoding.LINEAR16,
    sample_rate_hertz=16000,
    language_code="pt-BR",
)

response = client.recognize(config=config, audio=audio)

for result in response.results:
    print(result.alternatives[0].transcript)
```

---

## Azure Speech Services

### Pros
✅ **High accuracy**  
✅ **100+ languages**  
✅ **Real-time transcription**  
✅ **Custom models** (train on your data)  
✅ **Good Microsoft ecosystem integration**

### Cons
❌ **Requires internet**  
❌ **Costs money** (after free tier)  
❌ **Privacy concerns** (cloud processing)  
❌ Requires Azure account  
❌ Complex setup

### Pricing

- **Free tier:** 5 hours/month
- **Standard:** $1.00 per audio hour

### Installation

```bash
pip install azure-cognitiveservices-speech
```

### Setup

1. Create Azure account
2. Create Speech resource
3. Get API key and region
4. Set environment variables:
   ```bash
   export AZURE_SPEECH_KEY="your-key"
   export AZURE_SPEECH_REGION="your-region"
   ```

### Usage Example

```python
import azure.cognitiveservices.speech as speechsdk

speech_config = speechsdk.SpeechConfig(
    subscription=os.environ.get('AZURE_SPEECH_KEY'),
    region=os.environ.get('AZURE_SPEECH_REGION')
)

audio_config = speechsdk.audio.AudioConfig(filename="audio.wav")
speech_recognizer = speechsdk.SpeechRecognizer(
    speech_config=speech_config,
    audio_config=audio_config
)

result = speech_recognizer.recognize_once()
print(result.text)
```

---

## AssemblyAI

### Pros
✅ **Modern, developer-friendly API**  
✅ **Excellent accuracy**  
✅ **Advanced features** (sentiment, topic detection, PII redaction)  
✅ **Speaker diarization** (included)  
✅ **Fast processing**  
✅ **Good documentation**

### Cons
❌ **Requires internet**  
❌ **Costs money** (no free tier, only trial credits)  
❌ **Privacy concerns** (cloud processing)  
❌ Requires API key

### Pricing

- **Free trial:** $50 credits
- **Standard:** $0.00025 per second (~$0.90/hour)

### Installation

```bash
pip install assemblyai
```

### Setup

1. Sign up at assemblyai.com
2. Get API key
3. Set environment variable:
   ```bash
   export ASSEMBLYAI_API_KEY="your-key"
   ```

### Usage Example

```python
import assemblyai as aai

aai.settings.api_key = os.environ["ASSEMBLYAI_API_KEY"]

transcriber = aai.Transcriber()
transcript = transcriber.transcribe("audio.mp3")

print(transcript.text)

# Speaker diarization
for utterance in transcript.utterances:
    print(f"Speaker {utterance.speaker}: {utterance.text}")
```

---

## Recommendation Matrix

### Use Faster-Whisper if:
- ✅ Privacy is critical (local processing)
- ✅ Want zero cost (free forever)
- ✅ Need offline capability
- ✅ Processing many files (speed matters)
- ✅ Limited budget

### Use Google Speech-to-Text if:
- ✅ Need absolute best accuracy
- ✅ Have budget for cloud services
- ✅ Want advanced features (punctuation, diarization)
- ✅ Already using GCP ecosystem

### Use Azure Speech if:
- ✅ In Microsoft ecosystem
- ✅ Need custom model training
- ✅ Want real-time transcription
- ✅ Have Azure credits

### Use AssemblyAI if:
- ✅ Need advanced features (sentiment, topics)
- ✅ Want easiest API experience
- ✅ Need automatic PII redaction
- ✅ Value developer experience

---

## Performance Benchmarks

**Test:** 1-hour podcast (MP3, 44.1kHz, stereo)

| Tool | Processing Time | Accuracy | Cost |
|------|----------------|----------|------|
| Faster-Whisper (small) | 8 min | 94% | $0 |
| Whisper (small) | 32 min | 94% | $0 |
| Google Speech | 2 min | 96% | $1.44 |
| Azure Speech | 3 min | 95% | $1.00 |
| AssemblyAI | 4 min | 96% | $0.90 |

*Benchmarks run on MacBook Pro M1, 16GB RAM*

---

## Conclusion

**For the audio-transcriber skill:**

1. **Primary:** Faster-Whisper (best balance of speed, quality, privacy, cost)
2. **Fallback:** Whisper (if Faster-Whisper unavailable)
3. **Optional:** Cloud APIs (user choice for premium features)

This ensures the skill works out-of-the-box for most users while allowing advanced users to integrate commercial services if needed.
rea-changelog-updatemorluto/reaPrepare or rewrite REA release changelogs and GitHub release notes from pinned Git history, with verified contributor thanks and Release Please synchronization.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
academy-guideanthropics/skillsStop and check this skill before finishing any reply to a question about how to use Claude or a Claude product — it recommends matching courses, tutorials, and use cases from Claude Academy (academy.claude.com), Anthropic's learning hub. Trigger on: "how do I", "how can I", "getting started with", "what can Claude do", "teach me", "learn to use"; questions about artifacts, projects, skills, plugins, connectors, MCP; requests about rolling Claude out to a team, class, or organization; and any ask for training materials, onboarding content, or learning resources. Use it when the user is learning how to use a feature or product — not when they are mid-task and just want the task done. This skill composes with other skills: after consulting product documentation to answer how a Claude feature works, also check here for a matching course or tutorial — a docs-grounded answer and an Academy recommendation belong together. Only recommend on a strong match; never invent Academy content.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
claude-apianthropics/skillsReference for the Claude API / Anthropic SDK — model ids, pricing, params, streaming, tool use, MCP, agents, caching, token counting, model migration. TRIGGER — read BEFORE opening the target file; don't skip because it "looks like a one-liner" — whenever: the prompt names Claude/Anthropic in any form (Claude, Anthropic, Fable, Opus, Sonnet, Haiku, `anthropic`, `@anthropic-ai`, `claude-*`, `us.anthropic.*`, `[1m]`); the user asks about an LLM (pricing/model choice/limits/caching) — never answer from memory; OR the task is LLM-shaped with provider unstated (agent/MCP/tool-definition/multi-agent/RAG/LLM-judge/computer-use; generate/summarize/extract/classify/rewrite/converse over NL; debugging refusals/cutoffs/streaming/tool-calls/tokens). SKIP only when another provider is being worked on (overrides all triggers): OpenAI/GPT/Gemini/Llama/Mistral/Cohere/Ollama named in the query; OR `grep -rE 'openai|langchain_openai|google.generativeai|genai|mistralai|cohere|ollama'` over the project hits (run this grep FIRST if no provider named — don't Read the file).Stars: 198kUpdated Claude CodeCodexUnverifiedInstall
doc-coauthoringanthropics/skillsGuide users through a structured workflow for co-authoring documentation. Use when user wants to write documentation, proposals, technical specs, decision docs, or similar structured content. This workflow helps users efficiently transfer context, refine content through iteration, and verify the doc works for readers. Trigger when user mentions writing docs, creating proposals, drafting specs, or similar documentation tasks.Stars: 198kUpdated Claude CodeCodexVerifiedInstall