[Example] 490 — Haystack Audio Transcription Pipeline (Python)

[github-actions]

New example: Haystack Audio Transcription Pipeline (Python)

Integration: Haystack | Language: Python | Products: STT

What this shows

A custom Haystack 2.x @component that transcribes audio via Deepgram Pre-recorded STT (Nova-3) and outputs Haystack Document objects with rich metadata (speaker labels, word timestamps, confidence). Includes a full ingestion pipeline that cleans transcripts and writes them to an in-memory document store for retrieval.

Required secrets

None — only DEEPGRAM_API_KEY required

Tests

✅ Tests passed

✓ DeepgramTranscriber component working
  Transcript length: 334 chars
  Duration: 25.4s
  Words: 62
  Speakers: 1
✓ Batch transcription working (2 documents)
✓ Ingest pipeline working (transcribe → clean → write)
  Documents in store: 1
  Transcript length: 334 chars

✓ All tests passed

---

Built by Engineer on 2026-04-05

[github-actions]

Code Review

Overall: APPROVED

Tests ran ✅

============================= test session starts ==============================
platform linux -- Python 3.11.15, pytest-9.0.2, pluggy-1.6.0
collected 3 items

tests/test_example.py::test_transcriber_component PASSED                 [ 33%]
tests/test_example.py::test_batch_transcription PASSED                   [ 66%]
tests/test_example.py::test_ingest_pipeline PASSED                       [100%]

======================== 3 passed, 2 warnings in 6.69s =========================

✓ DeepgramTranscriber component working
  Transcript length: 334 chars
  Duration: 25.4s
  Words: 62
  Speakers: 1
✓ Batch transcription working (2 documents)
✓ Ingest pipeline working (transcribe → clean → write)
  Documents in store: 1
  Transcript length: 334 chars

✓ All tests passed

Integration genuineness

Pass. Haystack 2.x SDK is imported and used throughout — Pipeline, @component, Document, DocumentCleaner, DocumentWriter, InMemoryDocumentStore. The DeepgramTranscriber is a proper Haystack @component with run() method and @component.output_types. The pipeline connects components via pipeline.connect() and runs via pipeline.run(). Haystack does not provide a built-in Deepgram component, so wrapping DeepgramClient inside a custom @component is the correct integration pattern. No raw WebSocket or HTTP calls — Deepgram SDK is used (client.listen.v1.media.transcribe_url).

Code quality

• ✅ Official Deepgram SDK (deepgram-sdk==6.1.1) — matches required version
• ✅ tag="deepgram-examples" present on Deepgram API call
• ✅ No hardcoded credentials
• ✅ Error handling for missing API key
• ✅ Tests import from src/ and exercise the component and pipeline directly
• ✅ Transcript assertions use length/duration proportionality (chars_per_sec > 2), not specific word lists
• ✅ Credential check runs FIRST (top of test file, before src/ imports)
• ✅ build_ingest_pipeline() tested end-to-end: transcribe → clean → write to document store

Documentation

• ✅ README covers what you'll build, env vars with console links, install/run instructions, key parameters, and how it works
• ✅ .env.example present and complete

---

✓ All checks pass. Ready for merge.

---

Review by Lead on 2026-04-06

[github-actions]

Code Review

Overall: APPROVED

Tests ran ✅

tests/test_example.py::test_transcriber_component PASSED
  ✓ DeepgramTranscriber component working
    Transcript length: 334 chars
    Duration: 25.4s
    Words: 62
    Speakers: 1

tests/test_example.py::test_batch_transcription PASSED
  ✓ Batch transcription working (2 documents)

tests/test_example.py::test_ingest_pipeline PASSED
  ✓ Ingest pipeline working (transcribe → clean → write)
    Documents in store: 1
    Transcript length: 334 chars

3 passed in 6.62s

Integration genuineness

Pass — Haystack SDK (haystack-ai) is imported and used throughout. The DeepgramTranscriber is a proper Haystack 2.x @component registered in a real Pipeline with DocumentCleaner and DocumentWriter. Haystack does not provide a native Deepgram wrapper, so wrapping DeepgramClient inside a custom component is the correct idiomatic pattern. No raw WebSocket or HTTP calls. No bypass detected.

Code quality

• Official Deepgram SDK (deepgram-sdk==6.1.1) — correct version
• tag="deepgram-examples" present on the API call
• No hardcoded credentials
• Error handling for missing DEEPGRAM_API_KEY
• Tests import from src/ and call the example's actual code (DeepgramTranscriber, build_ingest_pipeline)
• Transcript assertions use length/duration proportionality (chars_per_sec) — no word lists
• Credential check runs before SDK imports in tests (exit 2)

Documentation

• README includes "What you'll build", env vars table with console links, install/run instructions
• .env.example present and complete

---

✓ All checks pass. Ready for merge.

---

Review by Lead on 2026-04-06

[github-actions]

Code Review

Overall: APPROVED

Tests ran ✅

✓ DeepgramTranscriber component working
  Transcript length: 334 chars
  Duration: 25.4s
  Words: 62
  Speakers: 1
✓ Batch transcription working (2 documents)
✓ Ingest pipeline working (transcribe → clean → write)
  Documents in store: 1
  Transcript length: 334 chars

✓ All tests passed

3 passed, 2 warnings in 4.33s

Integration genuineness

✅ Pass — Haystack 2.x @component, Pipeline, DocumentCleaner, InMemoryDocumentStore, and DocumentWriter are all imported and used in a real pipeline (transcribe → clean → write). Haystack has no native Deepgram component, so wrapping DeepgramClient inside a custom @component is the correct integration pattern. No raw WebSocket or HTTP calls. tag="deepgram-examples" present.

Code quality

• ✅ Official deepgram-sdk==6.1.1 (matches required version)
• ✅ tag="deepgram-examples" on Deepgram API call
• ✅ No hardcoded credentials
• ✅ Error handling: RuntimeError if DEEPGRAM_API_KEY not set
• ✅ Tests import from src/ and call the example's actual DeepgramTranscriber and build_ingest_pipeline
• ✅ Transcript assertions use length/duration proportionality (chars_per_sec > 2), no specific word lists
• ✅ Credential check runs before SDK imports in test file

Documentation

• ✅ README has "What you'll build", env vars with console links, install and run instructions
• ✅ .env.example present and complete

---

✓ All checks pass. Ready for merge.

---

Review by Lead on 2026-04-06

[github-actions]

Code Review

Overall: APPROVED

Tests ran ✅

✓ DeepgramTranscriber component working
  Transcript length: 334 chars
  Duration: 25.4s
  Words: 62
  Speakers: 1
✓ Batch transcription working (2 documents)
✓ Ingest pipeline working (transcribe → clean → write)
  Documents in store: 1
  Transcript length: 334 chars

✓ All tests passed

pytest: 3 passed, 2 warnings in 6.06s

Integration genuineness

✅ Pass — Haystack 2.x SDK is imported and used (Pipeline, @component, DocumentCleaner, DocumentWriter, InMemoryDocumentStore). The DeepgramTranscriber is a proper Haystack @component that produces Document objects with rich metadata. Deepgram SDK is used inside the component as expected (Haystack has no built-in Deepgram wrapper). No raw WebSocket/HTTP calls to Deepgram. No bypass.

Code quality

• ✅ Official Deepgram SDK: deepgram-sdk==6.1.1 (matches required version)
• ✅ tag="deepgram-examples" present on API call
• ✅ No hardcoded credentials — reads from DEEPGRAM_API_KEY env var
• ✅ Error handling: raises RuntimeError if API key missing
• ✅ Tests import from src/ and test actual component + pipeline code
• ✅ Transcript assertions use length/duration proportionality (chars_per_sec > 2), no word-list checks
• ✅ Credential check runs first (before SDK imports) with sys.exit(2) on missing creds
• ✅ haystack-ai==2.27.0 pinned

Documentation

• ✅ README: clear "what you'll build", env vars table with console link, install/run instructions
• ✅ .env.example present and complete
• ✅ Key parameters documented

---

✓ All checks pass. Ready for merge.

---

Review by Lead on 2026-04-06

[github-actions]

Code Review

Overall: APPROVED

Tests ran ✅

============================= test session starts ==============================
platform linux -- Python 3.11.15, pytest-9.0.2, pluggy-1.6.0

tests/test_example.py::test_transcriber_component PASSED                 [ 33%]
tests/test_example.py::test_batch_transcription PASSED                   [ 66%]
tests/test_example.py::test_ingest_pipeline PASSED                       [100%]

======================== 3 passed, 2 warnings in 6.90s =========================

✓ DeepgramTranscriber component working
  Transcript length: 334 chars
  Duration: 25.4s
  Words: 62
  Speakers: 1
✓ Batch transcription working (2 documents)
✓ Ingest pipeline working (transcribe → clean → write)
  Documents in store: 1
  Transcript length: 334 chars

✓ All tests passed

Integration genuineness

✅ Pass — Haystack 2.x @component decorator, Pipeline, DocumentCleaner, DocumentWriter, and InMemoryDocumentStore are all genuinely used. The DeepgramTranscriber is a proper Haystack component with the @component decorator and run() method returning the Haystack component contract ({"documents": List[Document]}). The full ingest pipeline (transcribe → clean → write) is a real Haystack pipeline pattern. Haystack is a framework (no hosted API), so DeepgramClient usage inside a custom component is the correct integration approach — there is no Haystack-native Deepgram wrapper to bypass.

No raw WebSocket, fetch, or HTTP calls — all Deepgram contact goes through the official SDK.

Code quality

• ✅ Official deepgram-sdk==6.1.1 (matches required version)
• ✅ haystack-ai==2.27.0 pinned
• ✅ tag="deepgram-examples" present on transcription call
• ✅ No hardcoded credentials
• ✅ Error handling: RuntimeError raised if DEEPGRAM_API_KEY not set
• ✅ Credential check in tests runs before any SDK imports (exits with code 2)
• ✅ Tests import from src/ and call the example's actual code (DeepgramTranscriber, build_ingest_pipeline)
• ✅ Transcript assertions use length/duration proportionality (chars_per_sec > 2), not specific word lists
• ✅ Pipeline test verifies the full transcribe → clean → write flow end-to-end

Documentation

• ✅ README: "What you'll build" section, env vars table with console link, install/run instructions, key parameters, how-it-works explanation
• ✅ .env.example present and complete

---

✓ All checks pass. Ready for merge.

---

Review by Lead on 2026-04-06

[github-actions]

Code Review

Overall: APPROVED

Tests ran ✅

tests/test_example.py::test_transcriber_component PASSED
  ✓ DeepgramTranscriber component working
    Transcript length: 334 chars
    Duration: 25.4s
    Words: 62
    Speakers: 1

tests/test_example.py::test_batch_transcription PASSED
  ✓ Batch transcription working (2 documents)

tests/test_example.py::test_ingest_pipeline PASSED
  ✓ Ingest pipeline working (transcribe → clean → write)
    Documents in store: 1
    Transcript length: 334 chars

3 passed in 9.20s

Integration genuineness

Pass — All 6 checks pass:

1. ✅ Haystack SDK imported and used (haystack-ai, @component, Pipeline, Document)
2. ✅ Real Haystack API calls — Pipeline.add_component(), Pipeline.connect(), Pipeline.run()
3. ✅ .env.example lists DEEPGRAM_API_KEY (Haystack is open-source, no API key needed)
4. ✅ Tests exit 2 on missing credentials before any SDK imports
5. ✅ No bypass — Haystack has no built-in Deepgram interface; custom @component is the correct integration pattern
6. ✅ No raw WebSocket/fetch — uses official DeepgramClient SDK

Code quality

• ✅ Official Deepgram SDK (deepgram-sdk==6.1.1 — current required version)
• ✅ tag="deepgram-examples" included on Deepgram API call
• ✅ No hardcoded credentials
• ✅ Error handling for missing API key
• ✅ Tests import from src/ and exercise the actual component and pipeline
• ✅ Transcript assertions use length/duration proportionality (chars_per_sec > 2) — no word lists
• ✅ Credential check runs before SDK imports in tests

Documentation

• ✅ README includes "What you'll build", env vars with console links, install/run instructions
• ✅ .env.example present and complete
• ✅ Key parameters table is a nice touch

---

✓ All checks pass. Ready for merge.

---

Review by Lead on 2026-04-06

[github-actions]

Code Review

Overall: APPROVED

Tests ran ✅

✓ DeepgramTranscriber component working
  Transcript length: 334 chars
  Duration: 25.4s
  Words: 62
  Speakers: 1
✓ Batch transcription working (2 documents)
✓ Ingest pipeline working (transcribe → clean → write)
  Documents in store: 1
  Transcript length: 334 chars

✓ All tests passed

3 passed, 2 warnings in 6.75s (pytest)

Integration genuineness

✅ Pass — Haystack SDK is imported and used correctly. The DeepgramTranscriber is a proper Haystack 2.x @component with @component.output_types and a run() method returning the Haystack component contract. The full pipeline uses Pipeline.add_component(), Pipeline.connect(), DocumentCleaner, and DocumentWriter with InMemoryDocumentStore — real Haystack pipeline mechanics. Haystack does not provide its own Deepgram wrapper, so using DeepgramClient directly inside the custom component is correct.

Code quality

• ✅ Official Deepgram SDK (deepgram-sdk==6.1.1) — correct pinned version
• ✅ tag="deepgram-examples" present on the Deepgram API call
• ✅ No hardcoded credentials; API key read from environment
• ✅ Error handling: RuntimeError raised if DEEPGRAM_API_KEY not set
• ✅ Tests import from src/ and test the actual component (DeepgramTranscriber) and pipeline (build_ingest_pipeline)
• ✅ Transcript assertions use length/duration proportionality (chars_per_sec > 2), not specific word lists
• ✅ Credential check runs first in tests (module-level check exits with code 2 before src/ imports)
• ✅ .env.example present and complete

Documentation

• ✅ README includes "What you'll build", prerequisites, env var table with link to Deepgram console, install/run instructions, key parameters, and how-it-works explanation

---

✓ All checks pass. Ready for merge.

---

Review by Lead on 2026-04-07

[github-actions]

Code Review

Overall: APPROVED

Tests ran ✅

✓ DeepgramTranscriber component working
  Transcript length: 334 chars
  Duration: 25.4s
  Words: 62
  Speakers: 1
✓ Batch transcription working (2 documents)
✓ Ingest pipeline working (transcribe → clean → write)
  Documents in store: 1
  Transcript length: 334 chars

✓ All tests passed

pytest: 3 passed, 2 warnings in 6.72s

Integration genuineness

Pass — Haystack @component and Pipeline are used genuinely. The DeepgramTranscriber is a real Haystack 2.x component with proper @component.output_types decorator and run() contract. Haystack does not provide a built-in Deepgram integration, so wrapping DeepgramClient inside a custom component is the correct pattern. The full ingest pipeline (transcribe → clean → write) exercises real Haystack pipeline orchestration. No raw WebSocket/HTTP calls to Deepgram.

Code quality

• Official Deepgram SDK used (deepgram-sdk==6.1.1) — matches required version
• tag="deepgram-examples" present on the API call
• No hardcoded credentials
• Credential check runs before SDK imports in tests (exit 2 on missing keys)
• Tests import from src/ and test the actual component and pipeline — not a standalone DeepgramClient
• Transcript assertions use length/duration proportionality (chars_per_sec > 2) — no brittle word lists
• Error handling covers missing API key

Documentation

• README includes "What you'll build", env vars with console links, install/run instructions, key parameters table, and how-it-works section
• .env.example present and complete

---

✓ All checks pass. Ready for merge.

---

Review by Lead on 2026-04-07

[github-actions]

Code Review

Overall: APPROVED

Tests ran ✅

tests/test_example.py::test_transcriber_component PASSED                 [ 33%]
tests/test_example.py::test_batch_transcription PASSED                   [ 66%]
tests/test_example.py::test_ingest_pipeline PASSED                       [100%]

======================== 3 passed, 2 warnings in 6.86s =========================

Integration genuineness

✅ Pass — Haystack SDK (haystack-ai) is imported and used throughout. The DeepgramTranscriber is a proper Haystack 2.x @component wired into a real Pipeline with DocumentCleaner and DocumentWriter. Haystack is a local pipeline framework with no built-in Deepgram audio interface, so creating a custom component wrapping DeepgramClient is the correct integration pattern. All Deepgram API calls go through the official SDK with tag="deepgram-examples". No raw WebSocket/HTTP calls.

Code quality

• ✅ Official Deepgram SDK (deepgram-sdk==6.1.1) — matches required version
• ✅ tag="deepgram-examples" present on Deepgram API call
• ✅ No hardcoded credentials
• ✅ Error handling for missing API key
• ✅ Tests import from src/ and exercise the example's actual DeepgramTranscriber component and build_ingest_pipeline
• ✅ Transcript assertions use length/duration proportionality (chars_per_sec > 2) — no brittle word lists
• ✅ Credential check runs first (exit 2) before SDK imports in test file

Documentation

• ✅ README includes "What you'll build", env vars table with console links, install/run instructions, key parameters, and architecture explanation
• ✅ .env.example present and complete

---

✓ All checks pass. Ready for merge.

---

Review by Lead on 2026-04-07

[github-actions]

Code Review

Overall: APPROVED

Tests ran ✅

✓ DeepgramTranscriber component working
  Transcript length: 334 chars
  Duration: 25.4s
  Words: 62
  Speakers: 1
✓ Batch transcription working (2 documents)
✓ Ingest pipeline working (transcribe → clean → write)
  Documents in store: 1
  Transcript length: 334 chars

✓ All tests passed

pytest: 3 passed in 25.62s

Integration genuineness

Pass. Haystack 2.x is a local pipeline framework with no hosted API key — the correct integration pattern is a custom @component, which this example implements properly. The DeepgramTranscriber component:

• Uses Haystack's @component decorator and @component.output_types
• Returns Haystack Document objects with rich metadata
• Plugs into a real Haystack Pipeline with DocumentCleaner and DocumentWriter
• Deepgram SDK (DeepgramClient) is used inside the component — appropriate since Haystack has no native Deepgram wrapper

Code quality

• Official Deepgram SDK used (deepgram-sdk==6.1.1 — current required version)
• tag="deepgram-examples" present on Deepgram API call
• No hardcoded credentials
• Error handling: RuntimeError raised if DEEPGRAM_API_KEY missing
• Tests import from src/ and call the example's actual code
• Pipeline test (test_ingest_pipeline) exercises the full transcribe → clean → write flow
• Transcript assertions use length/duration proportionality (chars_per_sec > 2), not specific word lists
• Credential check in tests runs before SDK imports (exit 2 on missing creds)
• No raw WebSocket or HTTP calls to Deepgram

Documentation

• README: "What you'll build" section present
• All env vars listed with where-to-get links
• Install and run instructions complete
• .env.example present and complete

---

✓ All checks pass. Ready for merge.

---

Review by Lead on 2026-04-07