Manipur Audio Tape Case: What the NFSU Forensic Examination Really Found — And Why the Evidence Is Contested
Two forensic laboratories examined recordings allegedly linking former Manipur Chief Minister N. Biren Singh to the 2023 ethnic violence — and reached very different conclusions. Here is what the primary record actually shows, and what it does not.
On August 7, 2026, the Additional Solicitor General of India told the Supreme Court that audio recordings allegedly implicating former Manipur Chief Minister N. Biren Singh in the state's 2023 ethnic violence were "clearly manipulated." Advocate Prashant Bhushan, for the petitioner, told the same bench, on the same day, that a different forensic laboratory had found a 93% probability that the voice on the tape was Singh's.
Both statements can be true at once. That is the central, and most misunderstood, feature of this case.
Forensic audio examination is not one test — it is several distinct questions that are routinely collapsed into a single headline. A recording can genuinely contain a real person's voice and still be inadmissible because of how it was handled afterward. It can also show clear signs of alteration while the underlying voice question remains unresolved.
Does the voice belong to a specific, named person?
Is the file the original, unaltered recording?
Has the file been edited, spliced, or otherwise modified?
Does it accurately represent what was said, in the order it was said?
Can it be relied upon as evidence in court?
This article does not treat the widely circulated figure of "41 aberrations" as an established forensic finding until its actual source is examined in §6. Readers are urged to draw the same distinction the Supreme Court itself has pressed the parties on: the difference between what a forensic report says and what a lawyer says about a forensic report in open court.
The case arises from the ethnic violence between the Meitei and Kuki-Zo communities that began in Manipur in May 2023, killing more than 200 people and displacing tens of thousands, per contemporaneous reporting. In August 2024, The Wire published clips — reportedly recorded without consent by a whistleblower at official premises associated with the then Chief Minister — in which a voice said to be Biren Singh's discusses the state's role in the violence, including a claimed conversation with Union Home Minister Amit Shah about the use of "bombs," and admissions relating to the looting of state armouries.
The Kuki Organization for Human Rights Trust (KOHUR), represented by Advocate Prashant Bhushan, filed a writ petition before the Supreme Court in November 2024 seeking a court-monitored, independent investigation. The petition initially anchored its claims to a roughly 48-minute clip; a longer version, running over two hours, was placed before the Court later in the proceedings. The Union and Manipur governments oppose the petition; Manipur's government has separately argued that KOHUR carries "ideological baggage" and a "separatist mindset" — a characterisation the petitioner disputes.
Sources: LiveLaw case archive · Bar and Bench · The Wire, "Manipur Tapes" series
| Date | Event | Forensic Significance |
|---|---|---|
| Aug 2024 | The Wire publishes leaked audio clips in parts (Aug 7 & Aug 20) | Origin of the disputed recording enters the public domain |
| Nov 8, 2024 | KOHUR's writ petition heard; SG Tushar Mehta says material sent to CFSL | First judicial engagement with the forensic question |
| Feb 3, 2025 | Bench of CJI Sanjiv Khanna & Justice Sanjay Kumar orders CFSL Guwahati report in sealed cover; Bhushan first cites the Truth Labs 93% figure | First forensic report sought; competing lab finding placed on record |
| ~May 2025 | Bench, dissatisfied with CFSL's report, directs a fresh FSL report | CFSL's initial findings considered inconclusive |
| Aug 2025 | Court directs NFSU, Gandhinagar to take up the examination | Matter formally moves to NFSU |
| Nov 3, 2025 | NFSU tells Court the clip is "tampered," not fit for voice comparison | First NFSU finding: authenticity in doubt |
| Jan 2026 | Court orders exam of the entire 48-min clip; NFSU again cannot compare; Bhushan asked to pursue an unaltered original | Court explicitly seeks a complete, not excerpted, recording |
| Apr 2026 | Parties confirm audio was modified to strip the recordist's own voice; petitioner submits longer clip (2 hrs 26 min+) | Modification conceded — but of one specific portion, not necessarily the substantive content |
| Apr 30, 2026 | Court directs the first-generation copy be sent for forensic exam | Court prioritises earliest available generation of the file |
| Jul 30, 2026 | NFSU's six-page report completed, authored by Dr Surbhi Mathur | Core document at the centre of the Aug 7 hearing |
| Aug 7, 2026 | ASG Bhati places NFSU report in sealed cover, seeks dismissal; Bhushan cites Truth Labs 93%; Court orders copies shared, confidentiality maintained | Latest finding on record — report itself remains sealed |
- Two separate NFSU examinations are on record. The first (Aug 2025) worked with the ~48-minute clip and reported "tampered/unfit" in November 2025. A second round followed the Court's April 2026 order to obtain the first-generation copy of the longer (2 hr 26 min+) recording, producing the six-page report dated July 30, 2026.
- The Court repeatedly pushed for the earliest-generation file — a materially important detail, since copies and re-recordings can introduce artefacts that mimic or mask genuine tampering signatures.
- Reference/admitted voice samples of Singh were separately sought from the State for comparison.
What NFSU's July 30, 2026 report itself states in full is not publicly available. The Supreme Court has ordered it remain confidential. What is known comes from ASG Bhati's characterisation of it in open court, as reported by multiple outlets — not from the document itself. Every account of its findings, including this one, is built on secondary reporting of oral submissions about a sealed document.
- The recordings — including the newer, extended clip — show multiple alterations.
- Even the first-generation copy had, per NFSU's conclusion as relayed to the Court, been altered, meaning no reliable forensic verification could be carried out.
- NFSU has not offered a conclusion on whether the voice belongs to Biren Singh — the finding is about recording integrity, not identity. "We could not verify" is scientifically different from "we verified it is not him."
- Earlier (2025) reporting indicated the clip was a truncated portion of a longer recording, was an edited version, and showed no indication of deepfake or AI-based editing at that stage.
Sources: Bar and Bench · LiveLaw · Asianet Newsable · The Tribune · ANI
VERIFIED
"NFSU found 41 abnormalities/aberrations in the audio."
During the August 7, 2026 hearing, ASG Aishwarya Bhati told the bench, referring to the newly submitted extended clip: "I spoke to the CFSL doctor and I told them to send me a letter and they have sent me a letter. Even this... had 41 alterations, I am told." Bar and Bench, The Tribune, ANI and Asianet Newsable independently report this in substantially the same terms.
- It was stated orally, in open court — this part is well corroborated across independent outlets.
- Its source, per the ASG herself, is a letter from a CFSL doctor — not the NFSU six-page report authored by Dr Surbhi Mathur.
- The CFSL letter has not been placed in the public domain and has not been independently reviewed by any media organisation covering the hearing, as far as could be verified.
- It is not defined in any source reviewed whether "41" means splice points, discontinuities, or another unit of measurement.
Truth Labs Forensic Services — described in court by Bhushan as founded by former CJI M.N. Venkatachaliah — examined material well before NFSU's involvement; its report was submitted to the Court alongside the petitioner's supplementary affidavit in early 2025.
- Reference samples S1–S6 were drawn from a press conference and media interview featuring Singh, sourced from public video; questioned samples were tagged Q2–Q4.
- Methodology combined auditory, spectrographic, statistical and likelihood analysis, applying a Chi-square statistical test.
- Stated conclusion: 93% probability that the male speaker in the questioned and reference recordings is the same person — described as a "high probability" match.
A Chi-square-based likelihood figure is a statistical statement about speaker similarity — not a fingerprint-grade identification, and not, by itself, a finding about whether the recording is unaltered, complete, or made in the alleged circumstances. Its reference samples came from public video rather than a controlled, forensically supervised session — a limitation worth noting. It has not been cross-verified by NFSU or CFSL.
Sources: The Wire · Maktoob Media · HW News English · The Hills Journal
Without asserting which factor actually applies here, forensic audio science recognises several reasons independent labs can diverge on ostensibly the same recording: different exhibits (Truth Labs' 2024–25 material vs. NFSU's later, distinct versions); different questions asked (similarity vs. integrity-plus-comparison); reference sample quality (public video vs. State-furnished admitted samples); generational loss from copies-of-copies; chain-of-custody differences given the whistleblower origin; and timing — Truth Labs' analysis predates NFSU's by over a year, and if the files are not identical, direct comparison is scientifically questionable regardless of either lab's competence.
"Voice match ≠ proof the recording is authentic.
Evidence of editing ≠ proof the voice is not genuine."The central forensic lesson of this case
Voice identification asks whether a voice is consistent with a known speaker — this is what Truth Labs' 93% figure addresses. Audio authentication asks whether the file itself has been altered from its original captured state — this is what NFSU's reported findings primarily address. A recording can contain a correctly identified, genuine voice and still have been selectively edited in ways that change its meaning. Conversely, a file can show clear signs of modification — such as the conceded removal of the recordist's own voice — while retained voice segments remain unaltered.
Forensic science distinguishes splicing (combining segments), selective deletion (removing portions, as conceded here to protect the recordist's identity), reordering, and routine compression/re-encoding from messaging apps or re-recording. None of these automatically implies the retained voice is fabricated.
Earlier (2025) NFSU reporting is described in one source as finding no indication of deepfake or AI-based editing. No source reviewed states NFSU's July 2026 report reached a different conclusion on synthetic voice generation. This article treats AI/deepfake manipulation as not an established finding on the current public record.
Chain of custody remains central to why this case has taken more than a year and a half to reach a forensic finding the Court considers workable: an anonymous whistleblower origin, media publication before laboratory submission, and multiple differing-length versions in circulation — each a point where hash values, metadata, and bitstream integrity could change.
The Court's repeated, pointed directions — for a complete rather than excerpted recording, a first-generation rather than downstream copy, admitted voice samples obtained specifically for comparison, and confidentiality of the sealed NFSU report — reflect standard judicial caution around expert evidence. The bench has not, on the public record, endorsed either laboratory's conclusion as final. This article offers no legal advice and does not predict how the Court will weigh the competing submissions.
- Preserve original evidence — first-generation files matter more than convenient copies.
- Examine complete recordings wherever possible, not party-selected excerpts.
- Maintain and document an unbroken chain of custody.
- Keep authenticity analysis analytically separate from speaker comparison.
- Record technical limitations honestly rather than forcing a conclusion.
- Document every transformation a piece of evidence undergoes.
- Preserve metadata at every stage of transfer.
- Use validated, defensible forensic methodologies.
- Report uncertainty as uncertainty — "cannot reliably compare" is a legitimate finding.
- Make conclusions proportionate to what the evidence actually supports.
This case is likely to be studied for what it shows about India's forensic audio infrastructure — CFSLs, State FSLs, NFSU, and digital forensic units generally — and the value of standardised protocols, methodological transparency between government and private labs, reproducibility, and accreditation standards that let courts weigh competing reports on a like-for-like basis. No specific policy reform has been publicly announced in response to this case.
General best practice, not a description of what was done here: a rigorous examination would secure the original device or earliest obtainable file generation; hash and preserve at the bitstream level before analysis; extract metadata; run time- and frequency-domain and spectrographic analysis; apply ENF analysis where feasible; examine compression history and splice points; check background-noise continuity; collect reference samples under controlled, supervised conditions rather than from public broadcast material; and seek independent verification by a second, unaffiliated laboratory before a conclusion is relied upon.
As of the most recent verified hearing (Aug 7, 2026): the Court has ordered NFSU's six-page report be furnished to both sides while remaining confidential. The Centre has urged dismissal of the petition; the petitioner disputes the Centre's characterisation, points to Truth Labs' 93% finding, and intends to share the NFSU report with Truth Labs for its response. No final judicial finding on authenticity or speaker identity has been made public. This article does not predict the outcome, and any claim that the case has been "decided" in either direction should be treated as inaccurate as of this writing.
The loudest question in this case — "Is it really him?" — is not, on the current record, the question either lab has definitively answered. The deeper questions are narrower: Is the recording before the Court an original or a downstream copy? Does the conceded alteration touch substantive content or only incidental material? Can a defensible speaker comparison be performed at all? Is the chain of custody sound enough to support a conclusion? Forensic science, done well, is supposed to narrow uncertainty — not manufacture false certainty in either direction.
| Claim | Source | Type | Verified? |
|---|---|---|---|
| Case title & number | LiveLaw · Verdictum | Secondary | Cross-confirmed |
| Aug 7 bench composition | LiveLaw · Bar and Bench · PTI · ANI | Secondary | 4 outlets |
| NFSU report: 6 pp., Jul 30, 2026, Dr Surbhi Mathur | Bar and Bench (quoting order) | Secondary, quoting order | Verified |
| "41 alterations" — CFSL doctor's letter | Bar and Bench · Tribune · ANI · Asianet | Secondary, oral submission | Courtroom statement verified; NFSU-report finding NOT verified |
| Truth Labs 93%, Chi-square, S1–S6/Q2–Q4 | The Wire · Maktoob · HW News · Hills Journal | Secondary, reviewed copy | Cross-confirmed |
| No deepfake/AI indication (2025 round) | Bar and Bench | Secondary | Single-source, lower confidence |
| Manipur toll (200+ killed, 70,000+ displaced) | The Wire | Secondary | Not checked against an official count |
The NFSU report (Jul 30, 2026) and the CFSL doctor's letter cited by ASG Bhati are both sealed or otherwise unavailable publicly. This article's account of their contents rests on secondary reporting of oral courtroom submissions, not direct review of the documents — a limitation especially relevant to the "41 aberrations" figure.
