Skip to content

Latest commit

Β 

History

History
160 lines (123 loc) Β· 6.43 KB

File metadata and controls

160 lines (123 loc) Β· 6.43 KB

Part 6. Audio Overview β€” Custom Strategies to Override the System Prompt

Core Problem: NotebookLM's Audio Overview has a strongly fixed internal system prompt β€” default two-host podcast format, predictable intro/outro patterns, excessively casual tone that may be inappropriate for academic contexts.


Strategy A: Embed Format Instructions in the Source Itself

Before generating audio, add a separate "Audio Script Guide" note as a source:

# [AUDIO_GUIDE] Audio Generation Rules

## Format
- Generate audio for this notebook as a single expert delivering a lecture.
- This is NOT a two-host conversation β€” it is a formal academic lecture by one speaker.
- Skip "Hello everyone" or "Today we're going to…" β€” begin directly with the central argument.

## Tone
- Academic but accessible; authoritative but not condescending
- Use concrete literary quotations and examples actively
- Define technical terms (e.g., 'anachrony,' 'free indirect discourse') on first use

## Structure
1. Central claim / Thesis statement (30 seconds)
2. Evidence 1 + literary example (2 minutes)
3. Evidence 2 + counter-reading (2 minutes)
4. Rebuttal of opposing interpretation (1 minute)
5. Conclusion + contemporary relevance (1 minute)

Strategy B: Override via Custom Instructions

This audio should follow the format of a celebrated university lecture.
Only ONE host/speaker appears β€” no dialogue between two hosts.
The lecturer opens by throwing out a provocative interpretive claim,
then handles anticipated student objections via internal self-questioning.
Include at least one passage of direct quotation from the primary source,
with brief interpretive commentary before and after the quotation.
Never end with a neat resolution β€” leave a productive critical tension open.

Strategy C: Induce TTS / Audiobook Mode

Read the uploaded source text aloud as a professional audiobook narrator would.
Do not add interpretation or personal commentary.
Read the original text exactly, but pause briefly between paragraphs.
Read stage directions or chapter headings in a clearly differentiated voice/register.
When textual footnotes appear, signal them: "Note begins… Note ends."

Best for: Original literary text recordings, self-study listening for EAL learners


Strategy D: Precision-Controlled Academic Debate Format

Two speakers appear, but with strictly defined intellectual positions:
- Speaker A: Formalist / New Critic (cites only internal textual evidence; 
  rejects biographical and historical context as fallacious)
- Speaker B: Historicist / New Historicist (foregrounds period context, 
  ideology, power relations in every interpretive move)

Rules:
1. No introductory pleasantries β€” begin immediately with the critical crux
2. Every claim must quote directly from the uploaded sources
3. Do NOT resolve into agreement β€” leave the critical tension unresolved
4. Close with an open question addressed to the listener as judge

Best for: Modelling critical debate for A-Level or university seminars


Strategy E: Multi-Genre Experimental Formats

The Trial Format:

Structure as a literary courtroom drama.
Prosecution (critical/ironic reading) vs Defence (sympathetic/authorial reading)
put a character on trial β€” e.g., "Is Jay Gatsby culpable for Myrtle's death?"
A Judge summarises at intervals. The Jury (listener) delivers the final verdict.

The Documentary Format:

Structure as a literary documentary: narrator + archival voice clips 
(simulated period voices or author quotations) + modern scholarly commentary.
Intercut 1920s American context with close textual analysis of Gatsby.

The Radio Drama Format:

Dramatise a key scene from the text with minimal narrator framing.
After the dramatisation, a single literary critic provides a 3-minute close reading.
End with one unanswered critical question for student discussion.

Audio Strategy Comparison

Strategy Format Optimal Use Note
A (Source insertion) Academic lecture Concept overview, revision May be diluted if many sources
B (Custom Instructions) Self-questioning lecture Argument structure modelling Override not always complete
C (TTS mode) Audiobook Original text listening, EAL No commentary β€” self-study only
D (Precision debate) Academic debate Critical thinking, seminar prep Must maintain unresolved tension
E (Genre experiment) Trial / documentary Engagement, motivation May sacrifice academic register

English Literature Audio Scenarios β€” 5 Applications

  1. Self-Study Text Recording: 15-min audio of original text + key vocabulary gloss only
  2. Commuter Revision Summary: All 5 key themes covered in a 10-minute compressed lecture
  3. Debate Model Audio: Ideal teacher-student dialogue modelling literary argument
  4. Comparative Analysis Lecture: Two texts quoted alternately in contrapuntal close reading
  5. Pre-Exam Summary: 5-minute high-priority revision audio for the night before

Recommended Audio Prompts by Text

Hamlet

Generate a 10-minute lecture audio on the "To be or not to be" soliloquy.
Analyse it across three critical lenses in sequence:
1. Humanist / Renaissance context (mortality, free will)
2. Psychoanalytic reading (Freudian death drive, repression)
3. Theatrical / performance studies (actor's choices, staging history)
End each section with a single unanswered question for student reflection.

The Great Gatsby

Generate an audio debate between:
- Speaker A: Gatsby as America's tragic idealist (sympathetic)
- Speaker B: Gatsby as moral bankrupt enabled by class mythology (critical)
Each speaker cites 3 specific passages from the novel.
Do not reach consensus. Close with the question: "Is the green light hope or delusion?"

Pride and Prejudice

Generate an audiobook-style reading of the opening chapter of Pride and Prejudice,
using distinct vocal registers for narrator voice (ironic, controlled)
and dialogue (differentiated by character social register).
After the reading, provide 5 minutes of close textual commentary
focused exclusively on Austen's use of free indirect discourse.

Practical Tip: After audio generation, use NotebookLM's provided timestamps to select a specific segment and follow up: "Explain the argument made between [2:30–4:00] in more depth" β€” creating a layered audio + text study workflow.