whisper

Name: whisper
Author: AlexAI-MCP

132stars24forks

How to install whisper?

alexai-mcp/hermes-ccc/whisper

$npx -y skills add AlexAI-MCP/hermes-CCC --skill whisper

Installs into the current project.

Summary

OpenAI Whisper for speech recognition and transcription — local inference, multiple model sizes, language detection, and subtitle generation.

Files · 1

View on GitHub

SKILL.md

1	# Whisper
2
3	## Purpose
4
5	- Use this skill for local speech-to-text transcription and subtitle generation.
6	- Prefer it when privacy, offline processing, or batch audio workflows matter.
7	- Whisper works well for meetings, podcasts, interviews, voice notes, and extracted video audio.
8	- It supports multilingual transcription and language detection.
9
10	## Install
11
12	- Standard Whisper package:
13
14	```bash
15	pip install openai-whisper
16	```
17
18	- Faster inference alternative:
19
20	```bash
21	pip install faster-whisper
22	```
23
24	- `faster-whisper` is often the better default for production batch jobs because it is typically faster at similar accuracy.
25
26	## Command Line Usage
27
28	- Basic CLI transcription:
29
30	```bash
31	whisper audio.mp3 --model medium --language en
32	```
33
34	- Use GPU explicitly:
35
36	```bash
37	whisper audio.mp3 --model medium --language en --device cuda
38	```
39
40	- Generate subtitles in SRT format:
41
42	```bash
43	whisper audio.mp3 --model medium --language en --output_format srt
44	```
45
46	- The CLI is a good fit for one-off transcription, shell scripts, and media preprocessing jobs.
47
48	## Python API
49
50	```python
51	import whisper
52
53	model = whisper.load_model("medium")
54	result = model.transcribe("audio.mp3")
55
56	print(result["text"])
57	print(result["language"])
58	print(result["segments"][:2])
59	```
60
61	- Standard load pattern:
62	- `import whisper`
63	- `model = whisper.load_model('medium')`
64
65	## What `transcribe()` Returns
66
67	- `text`: the full transcript
68	- `segments`: timestamped segment-level outputs
69	- `language`: detected or selected language code
70
71	- Example shape:
72
73	```python
74	{
75	"text": "Full transcript text",
76	"language": "en",
77	"segments": [
78	{"id": 0, "start": 0.0, "end": 4.5, "text": "Hello everyone"},
79	],
80	}
81	```
82
83	## Model Sizes
84
85	- `tiny`
86	- `base`
87	- `small`
88	- `medium`
89	- `large-v3`
90
91	- The tradeoff is simple:
92	- smaller models are faster and cheaper
93	- larger models are slower but more accurate
94
95	## Model Selection Guidance
96
97	- `tiny`: fast experiments and low-resource CPU runs
98	- `base`: simple automation on clean audio
99	- `small`: balanced for lightweight production tasks
100	- `medium`: common quality default for serious transcription
101	- `large-v3`: best accuracy when latency and VRAM are acceptable
102
103	## Language Selection
104
105	- Set the language when you know it:
106
107	```bash
108	whisper audio.mp3 --model medium --language en
109	```
110
111	- Explicit language hints usually improve speed and stability.
112	- If the language is unknown, let the model detect it.
113
114	## Language Detection
115
116	- Whisper can estimate the spoken language from audio features.
117	- Example pattern:
118
119	```python
120	import whisper
121
122	model = whisper.load_model("medium")
123	audio = whisper.load_audio("audio.mp3")
124	audio = whisper.pad_or_trim(audio)
125	mel = whisper.log_mel_spectrogram(audio).to(model.device)
126
127	_, probs = model.detect_language(mel)
128	language = max(probs, key=probs.get)
129	print(language)
130	```
131
132	- `model.detect_language(audio)` is the key workflow concept, though in practice you pass the processed spectrogram tensor.
133
134	## Subtitle Generation
135
136	- Generate `.srt` subtitles from the CLI:
137
138	```bash
139	whisper audio.mp3 --model medium --language en --output_format srt
140	```
141
142	- Subtitle outputs are useful for:
143	- video captions
144	- podcast transcripts
145	- lecture indexing
146	- searchable archives
147
148	## Batch Processing Multiple Files
149
150	- Simple shell loop:
151
152	```bash
153	Get-ChildItem *.mp3 \| ForEach-Object {
154	whisper $_.FullName --model medium --language en --output_format srt
155	}
156	```
157
158	- Python batch example:
159
160	```python
161	from pathlib import Path
162
163	import whisper
164
165	model = whisper.load_model("medium")
166
167	for path in Path("audio").glob("*.mp3"):
168	result = model.transcribe(str(path))
169	out_path = path.with_suffix(".txt")
170	out_path.write_text(result["text"], encoding="utf-8")
171	```
172
173	- Batch processing is a common pattern for meeting folders, call archives, and downloaded media collections.
174
175	## GPU vs CPU
176
177	- GPU example from the CLI:
178
179	```bash
180	whisper audio.mp3 --model medium --device cuda
181	```
182
183	- GPU is strongly preferred for:
184	- `medium`
185	- `large-v3`
186	- multi-file batch jobs
187
188	- CPU is acceptable for:
189	- `tiny`
190	- `base`
191	- occasional short clips
192
193	## Common Use Cases
194
195	- YouTube audio transcription after extracting audio from video
196	- meeting notes from Zoom or Teams recordings
197	- podcast transcription for search and republishing
198	- lecture indexing and subtitle generation
199	- multilingual voice note transcription
200
201	## `faster-whisper`
202
203	- Install with:
204
205	```bash
206	pip install faster-whisper
207	```
208
209	- It is often around 4x faster while maintaining comparable accuracy.
210	- It is a strong choice for MLOps pipelines where throughput matters.
211
212	## `faster-whisper` Example
213
214	```python
215	from faster_whisper import WhisperModel
216
217	m

People who liked this also viewed

6 picks

Type

find-skillsHelps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express interest in extending capabilities. This skSkillsJul 20262.6M27k
grill-meA relentless interview to sharpen a plan or design.SkillsJul 2026633k182k
vercel-react-best-practicesVercel's official collection of agent skillsSkillsJul 2026573k29k
improve-codebase-architectureScan a codebase for deepening opportunities, present them as a visual HTML report, then grill through whichever one you pick.SkillsJul 2026522k182k
setup-matt-pocock-skillsConfigure this repo for the engineering skills — set up its issue tracker, triage label vocabulary, and domain doc layout. Run once before first use of the other engineering skills.SkillsJul 2026454k182k
triageMove issues and external PRs through a state machine of triage roles — categorise, verify, grill if needed, and write agent-ready briefs.SkillsJul 2026424k182k

whisper

132stars24forks

View on GitHub

How to install whisper?

alexai-mcp/hermes-ccc/whisper

$npx -y skills add AlexAI-MCP/hermes-CCC --skill whisper

Installs into the current project.

Summary

OpenAI Whisper for speech recognition and transcription — local inference, multiple model sizes, language detection, and subtitle generation.

Files · 1

View on GitHub

SKILL.md

1	# Whisper
2
3	## Purpose
4
5	- Use this skill for local speech-to-text transcription and subtitle generation.
6	- Prefer it when privacy, offline processing, or batch audio workflows matter.
7	- Whisper works well for meetings, podcasts, interviews, voice notes, and extracted video audio.
8	- It supports multilingual transcription and language detection.
9
10	## Install
11
12	- Standard Whisper package:
13
14	```bash
15	pip install openai-whisper
16	```
17
18	- Faster inference alternative:
19
20	```bash
21	pip install faster-whisper
22	```
23
24	- `faster-whisper` is often the better default for production batch jobs because it is typically faster at similar accuracy.
25
26	## Command Line Usage
27
28	- Basic CLI transcription:
29
30	```bash
31	whisper audio.mp3 --model medium --language en
32	```
33
34	- Use GPU explicitly:
35
36	```bash
37	whisper audio.mp3 --model medium --language en --device cuda
38	```
39
40	- Generate subtitles in SRT format:
41
42	```bash
43	whisper audio.mp3 --model medium --language en --output_format srt
44	```
45
46	- The CLI is a good fit for one-off transcription, shell scripts, and media preprocessing jobs.
47
48	## Python API
49
50	```python
51	import whisper
52
53	model = whisper.load_model("medium")
54	result = model.transcribe("audio.mp3")
55
56	print(result["text"])
57	print(result["language"])
58	print(result["segments"][:2])
59	```
60
61	- Standard load pattern:
62	- `import whisper`
63	- `model = whisper.load_model('medium')`
64
65	## What `transcribe()` Returns
66
67	- `text`: the full transcript
68	- `segments`: timestamped segment-level outputs
69	- `language`: detected or selected language code
70
71	- Example shape:
72
73	```python
74	{
75	"text": "Full transcript text",
76	"language": "en",
77	"segments": [
78	{"id": 0, "start": 0.0, "end": 4.5, "text": "Hello everyone"},
79	],
80	}
81	```
82
83	## Model Sizes
84
85	- `tiny`
86	- `base`
87	- `small`
88	- `medium`
89	- `large-v3`
90
91	- The tradeoff is simple:
92	- smaller models are faster and cheaper
93	- larger models are slower but more accurate
94
95	## Model Selection Guidance
96
97	- `tiny`: fast experiments and low-resource CPU runs
98	- `base`: simple automation on clean audio
99	- `small`: balanced for lightweight production tasks
100	- `medium`: common quality default for serious transcription
101	- `large-v3`: best accuracy when latency and VRAM are acceptable
102
103	## Language Selection
104
105	- Set the language when you know it:
106
107	```bash
108	whisper audio.mp3 --model medium --language en
109	```
110
111	- Explicit language hints usually improve speed and stability.
112	- If the language is unknown, let the model detect it.
113
114	## Language Detection
115
116	- Whisper can estimate the spoken language from audio features.
117	- Example pattern:
118
119	```python
120	import whisper
121
122	model = whisper.load_model("medium")
123	audio = whisper.load_audio("audio.mp3")
124	audio = whisper.pad_or_trim(audio)
125	mel = whisper.log_mel_spectrogram(audio).to(model.device)
126
127	_, probs = model.detect_language(mel)
128	language = max(probs, key=probs.get)
129	print(language)
130	```
131
132	- `model.detect_language(audio)` is the key workflow concept, though in practice you pass the processed spectrogram tensor.
133
134	## Subtitle Generation
135
136	- Generate `.srt` subtitles from the CLI:
137
138	```bash
139	whisper audio.mp3 --model medium --language en --output_format srt
140	```
141
142	- Subtitle outputs are useful for:
143	- video captions
144	- podcast transcripts
145	- lecture indexing
146	- searchable archives
147
148	## Batch Processing Multiple Files
149
150	- Simple shell loop:
151
152	```bash
153	Get-ChildItem *.mp3 \| ForEach-Object {
154	whisper $_.FullName --model medium --language en --output_format srt
155	}
156	```
157
158	- Python batch example:
159
160	```python
161	from pathlib import Path
162
163	import whisper
164
165	model = whisper.load_model("medium")
166
167	for path in Path("audio").glob("*.mp3"):
168	result = model.transcribe(str(path))
169	out_path = path.with_suffix(".txt")
170	out_path.write_text(result["text"], encoding="utf-8")
171	```
172
173	- Batch processing is a common pattern for meeting folders, call archives, and downloaded media collections.
174
175	## GPU vs CPU
176
177	- GPU example from the CLI:
178
179	```bash
180	whisper audio.mp3 --model medium --device cuda
181	```
182
183	- GPU is strongly preferred for:
184	- `medium`
185	- `large-v3`
186	- multi-file batch jobs
187
188	- CPU is acceptable for:
189	- `tiny`
190	- `base`
191	- occasional short clips
192
193	## Common Use Cases
194
195	- YouTube audio transcription after extracting audio from video
196	- meeting notes from Zoom or Teams recordings
197	- podcast transcription for search and republishing
198	- lecture indexing and subtitle generation
199	- multilingual voice note transcription
200
201	## `faster-whisper`
202
203	- Install with:
204
205	```bash
206	pip install faster-whisper
207	```
208
209	- It is often around 4x faster while maintaining comparable accuracy.
210	- It is a strong choice for MLOps pipelines where throughput matters.
211
212	## `faster-whisper` Example
213
214	```python
215	from faster_whisper import WhisperModel
216
217	m

People who liked this also viewed

6 picks

Type