azure-speech-to-text

Name: azure-speech-to-text
Author: calesthio

46installs4.9kforks

How to install azure-speech-to-text?

calesthio/openmontage/azure-speech-to-text

$npx -y skills add calesthio/OpenMontage --skill azure-speech-to-text

Installs into the current project.

Summary

Transcribe audio to text using Azure AI Speech (Fast Transcription REST API). Use when converting audio/video to text, generating subtitles, or processing spoken content in OpenMontage. Optional cloud STT provider — preferred when AZURE_SPEECH_KEY is configured; the local faster-

Files · 1

View on GitHub

SKILL.md

1	# Azure AI Speech — Speech-to-Text
2
3	Transcribe audio to text with Azure Fast Transcription — synchronous,
4	word-level timestamps, speaker diarization, and multi-language identification.
5	In OpenMontage this is exposed through the `azure_stt` tool (`capability=analysis`,
6	`provider=azure`). It is an optional cloud STT provider — when
7	`AZURE_SPEECH_KEY` is configured, prefer it for cloud transcription. The local
8	`transcriber` tool (faster-whisper) remains the default offline path and the
9	fallback when Azure is unavailable.
10
11	> Docs: [Fast Transcription](https://learn.microsoft.com/azure/ai-services/speech-service/fast-transcription-create) · [Speech service overview](https://learn.microsoft.com/azure/ai-services/speech-service/spx-overview)
12
13	## Why Fast Transcription (not Batch)
14
15	Azure exposes three STT surfaces. OpenMontage uses Fast Transcription because
16	the pipeline transcribes local audio files:
17
18	\| Surface \| Input \| Latency \| Needs \|
19	\|---------\|-------\|---------\|-------\|
20	\| Fast Transcription (used here) \| local file, multipart POST \| synchronous, sub-real-time \| key + region \|
21	\| Batch Transcription \| audio at a URL (Blob + SAS) \| async job + polling \| Blob storage plumbing \|
22	\| Speech SDK (`spx`) \| mic / stream / file \| streaming \| native `azure-cognitiveservices-speech` package \|
23
24	Fast Transcription needs no Blob storage, no SAS URLs, and no native SDK — just
25	`requests` and the two env vars.
26
27	## Setup
28
29	Create a Speech resource in the [Azure portal](https://portal.azure.com);
30	copy the key and region from its Keys and Endpoint page.
31
32	```bash
33	export AZURE_SPEECH_KEY=your_speech_resource_key
34	export AZURE_SPEECH_REGION=eastus # your resource's region
35	# export AZURE_SPEECH_ENDPOINT=https://... # optional: overrides region
36	```
37
38	`azure_stt` reports `AVAILABLE` once `AZURE_SPEECH_KEY` plus either
39	`AZURE_SPEECH_REGION` or `AZURE_SPEECH_ENDPOINT` are set.
40
41	## Using it in a pipeline
42
43	Prefer `azure_stt` over `transcriber` unless the run must be offline. Its output
44	matches the `transcriber` schema exactly, so it is a drop-in for `subtitle_gen`
45	and any stage that consumes a transcript.
46
47	```python
48	from tools.tool_registry import registry
49	registry.discover()
50	stt = registry._tools["azure_stt"]
51
52	result = stt.execute({
53	"input_path": "projects/my-video/assets/audio/narration.mp3",
54	# "language": "en", # ISO 639-1 or BCP-47 ("en-US"); omit for auto-ID
55	# "diarize": True, # speaker labels, no HuggingFace token needed
56	# "max_speakers": 4,
57	"output_dir": "projects/my-video/artifacts",
58	})
59	if result.success:
60	segs = result.data["segments"] # [{id,start,end,text,words:[...]}]
61	words = result.data["word_timestamps"] # flat [{word,start,end,probability}]
62	```
63
64	If `azure_stt` is unavailable (no key) or errors, fall back to `transcriber`
65	(local whisper) — its `execute` signature and output are identical.
66
67	## Parameters that matter
68
69	- `language` — pass an ISO code (`"en"`) or a full locale (`"en-US"`). Pin it
70	when you know the language; it is faster and more accurate than auto-ID.
71	- `candidate_locales` — when `language` is omitted, Azure runs language
72	identification across this shortlist. Narrow it to the languages you actually
73	expect; a huge list slows detection and invites misclassification.
74	- `diarize` / `max_speakers` — enable for multi-speaker audio (interviews,
75	podcasts). Set `max_speakers` to the real upper bound.
76	- `profanity_filter` — `None` \| `Masked` (default) \| `Removed` \| `Tags`.
77
78	## Response shape (mapped to the transcriber schema)
79
80	The raw Azure response (`phrases[]` with `offsetMilliseconds` / `words[]`) is
81	converted to seconds and the OpenMontage transcript schema:
82
83	```json
84	{
85	"segments": [
86	{"id": 0, "start": 0.0, "end": 2.4, "text": "Hello world",
87	"speaker": 1,
88	"words": [{"word": "Hello", "start": 0.0, "end": 0.5, "probability": 0.98}]}
89	],
90	"word_timestamps": [{"word": "Hello", "start": 0.0, "end": 0.5, "probability": 0.98}],
91	"language": "en-US",
92	"duration_seconds": 2.4,
93	"provider": "azure"
94	}
95	```
96
97	Note: Fast Transcription has no per-word confidence, so each word carries the
98	phrase confidence in `probability`.
99
100	## Limits & tips
101
102	- Single file up to ~2 hours / a few hundred MB per request. For longer or bulk
103	jobs, use Azure Batch Transcription instead.
104	- Send

People who liked this also viewed

6 picks

Type

azure-speech-to-text

46installs4.9kforks

View on GitHub

How to install azure-speech-to-text?

calesthio/openmontage/azure-speech-to-text

$npx -y skills add calesthio/OpenMontage --skill azure-speech-to-text

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Azure AI Speech — Speech-to-Text
2
3	Transcribe audio to text with Azure Fast Transcription — synchronous,
4	word-level timestamps, speaker diarization, and multi-language identification.
5	In OpenMontage this is exposed through the `azure_stt` tool (`capability=analysis`,
6	`provider=azure`). It is an optional cloud STT provider — when
7	`AZURE_SPEECH_KEY` is configured, prefer it for cloud transcription. The local
8	`transcriber` tool (faster-whisper) remains the default offline path and the
9	fallback when Azure is unavailable.
10
11	> Docs: [Fast Transcription](https://learn.microsoft.com/azure/ai-services/speech-service/fast-transcription-create) · [Speech service overview](https://learn.microsoft.com/azure/ai-services/speech-service/spx-overview)
12
13	## Why Fast Transcription (not Batch)
14
15	Azure exposes three STT surfaces. OpenMontage uses Fast Transcription because
16	the pipeline transcribes local audio files:
17
18	\| Surface \| Input \| Latency \| Needs \|
19	\|---------\|-------\|---------\|-------\|
20	\| Fast Transcription (used here) \| local file, multipart POST \| synchronous, sub-real-time \| key + region \|
21	\| Batch Transcription \| audio at a URL (Blob + SAS) \| async job + polling \| Blob storage plumbing \|
22	\| Speech SDK (`spx`) \| mic / stream / file \| streaming \| native `azure-cognitiveservices-speech` package \|
23
24	Fast Transcription needs no Blob storage, no SAS URLs, and no native SDK — just
25	`requests` and the two env vars.
26
27	## Setup
28
29	Create a Speech resource in the [Azure portal](https://portal.azure.com);
30	copy the key and region from its Keys and Endpoint page.
31
32	```bash
33	export AZURE_SPEECH_KEY=your_speech_resource_key
34	export AZURE_SPEECH_REGION=eastus # your resource's region
35	# export AZURE_SPEECH_ENDPOINT=https://... # optional: overrides region
36	```
37
38	`azure_stt` reports `AVAILABLE` once `AZURE_SPEECH_KEY` plus either
39	`AZURE_SPEECH_REGION` or `AZURE_SPEECH_ENDPOINT` are set.
40
41	## Using it in a pipeline
42
43	Prefer `azure_stt` over `transcriber` unless the run must be offline. Its output
44	matches the `transcriber` schema exactly, so it is a drop-in for `subtitle_gen`
45	and any stage that consumes a transcript.
46
47	```python
48	from tools.tool_registry import registry
49	registry.discover()
50	stt = registry._tools["azure_stt"]
51
52	result = stt.execute({
53	"input_path": "projects/my-video/assets/audio/narration.mp3",
54	# "language": "en", # ISO 639-1 or BCP-47 ("en-US"); omit for auto-ID
55	# "diarize": True, # speaker labels, no HuggingFace token needed
56	# "max_speakers": 4,
57	"output_dir": "projects/my-video/artifacts",
58	})
59	if result.success:
60	segs = result.data["segments"] # [{id,start,end,text,words:[...]}]
61	words = result.data["word_timestamps"] # flat [{word,start,end,probability}]
62	```
63
64	If `azure_stt` is unavailable (no key) or errors, fall back to `transcriber`
65	(local whisper) — its `execute` signature and output are identical.
66
67	## Parameters that matter
68
69	- `language` — pass an ISO code (`"en"`) or a full locale (`"en-US"`). Pin it
70	when you know the language; it is faster and more accurate than auto-ID.
71	- `candidate_locales` — when `language` is omitted, Azure runs language
72	identification across this shortlist. Narrow it to the languages you actually
73	expect; a huge list slows detection and invites misclassification.
74	- `diarize` / `max_speakers` — enable for multi-speaker audio (interviews,
75	podcasts). Set `max_speakers` to the real upper bound.
76	- `profanity_filter` — `None` \| `Masked` (default) \| `Removed` \| `Tags`.
77
78	## Response shape (mapped to the transcriber schema)
79
80	The raw Azure response (`phrases[]` with `offsetMilliseconds` / `words[]`) is
81	converted to seconds and the OpenMontage transcript schema:
82
83	```json
84	{
85	"segments": [
86	{"id": 0, "start": 0.0, "end": 2.4, "text": "Hello world",
87	"speaker": 1,
88	"words": [{"word": "Hello", "start": 0.0, "end": 0.5, "probability": 0.98}]}
89	],
90	"word_timestamps": [{"word": "Hello", "start": 0.0, "end": 0.5, "probability": 0.98}],
91	"language": "en-US",
92	"duration_seconds": 2.4,
93	"provider": "azure"
94	}
95	```
96
97	Note: Fast Transcription has no per-word confidence, so each word carries the
98	phrase confidence in `probability`.
99
100	## Limits & tips
101
102	- Single file up to ~2 hours / a few hundred MB per request. For longer or bulk
103	jobs, use Azure Batch Transcription instead.
104	- Send

People who liked this also viewed

6 picks

Type