Turn Telegram voice notes into transcribed, structured records via n8n, with a glossary prompt to fix mangled proper nouns.
“Voice Note to Structured Record Pipeline with Glossary Correction” is an intermediate Operations AI agent workflow from Any. It connects n8n and ChatGPT in 5 steps. Setup takes about Not stated and running costs are about Not stated.
Call notes get retyped by hand after every conversation, which is slow and error-prone. A voice-note-to-record automation helps, but transcription mangles proper nouns like company names, so records look clean while being quietly wrong. A glossary prompt corrects most of these errors, but the pipeline still needs a way to catch bad proper nouns before they reach the record.
Send call notes as Telegram voice messages to a bot connected to your n8n workflow. This replaces typing notes after each call.
Add a transcription step in n8n that converts the voice note into text before any structuring happens.
Pass the transcript through an LLM step with a glossary of expected company and person names so mangled proper nouns are corrected before the record is built.
Map the corrected transcript into a structured record with the fields you need for your notes system.
Add a check that highlights names the model was unsure about so a human can confirm them instead of trusting a clean-looking but wrong record.
You are correcting a call transcript before it becomes a structured record. Here is the raw transcript: [TRANSCRIPT]. Here is my glossary of expected company and person names: [GLOSSARY]. Rewrite the transcript, replacing any mangled or phonetically similar proper nouns with the correct glossary entries. Do not change anything else. Then output a structured record with these fields: [FIELDS]. After the record, list every proper noun you corrected and every proper noun you were unsure about, so I can verify them manually.
Help me build an n8n workflow that takes a Telegram voice note, transcribes it, sends the transcript through a glossary correction prompt, and outputs a structured record. Use [TRANSCRIPTION SERVICE] for audio-to-text and [LLM] for correction and structuring. Include a step that flags proper nouns the model is uncertain about. Describe each node, its configuration, and how data passes between them. Do not invent credentials or services I have not listed.
Turn Telegram voice notes into transcribed, structured records via n8n, with a glossary prompt to fix mangled proper nouns.
You need n8n and ChatGPT.
Plan for about Not stated. It is rated intermediate. The guide walks through 5 steps, starting with “Capture voice notes in Telegram”.
Expect about Not stated in running costs (Under $25/mo range).
Call notes get retyped by hand after every conversation, which is slow and error-prone. A voice-note-to-record automation helps, but transcription mangles proper nouns like company names, so records look clean while being quietly wrong. A glossary prompt corrects most of these errors, but the pipeline still needs a way to catch bad proper nouns before they reach the record.
After each sales call, Zapier sends the transcript to Claude, which writes the notes and updates next steps and deal fields in Salesforce.
Score every demo request against your ICP, update HubSpot and ping the right rep in Slack within a minute of the form submit.
Every new Zendesk ticket gets a category, priority and sentiment tag plus a draft reply in an internal note before an agent opens it.
Built this workflow? Show it is featured on AgentUseHub with a live upvote count that links back here. About badges
