DataHackIL · shaypal5 · May 12, 2026 · May 11, 2026 · May 11, 2026
diff --git a/CLAUDE.md b/CLAUDE.md
@@ -51,15 +51,25 @@ Do not merge data without updating the delivery log.
 ### `has_violence` is a derived convenience field — keep it
 
 `has_violence` appears in `weak_label.has_violence` (clip `.json`) and as a column in
-`manifest.csv`. It is **derived** from the hierarchical taxonomy, not assigned independently:
+`manifest.csv`. It is **derived from the strong-label events**, not from typology or intensity.
+The authoritative rule lives in `synthbanshee/labels/generator.py` and is pinned in
+[spec §5.1](https://github.com/DataHackIL/SynthBanshee/blob/main/docs/spec.md#51-per-clip-metadata-json):
 
+```python
+has_violence = any(e.tier1_category != "NONE" for e in events)
 ```
-has_violence = (violence_typology in {"SV", "IT", "NEG"}) and (max_intensity >= 3)
-```
+
+Corollaries: empty `events` → `False`; `NEG` clips are correctly `False` (by §4.1 NEG is
+acoustically intense but non-violent, so every event lands `tier1_category: "NONE"`);
+`violence_typology` and `has_violence` may legitimately disagree (e.g. `SV` typology with
+`has_violence: False` if no event carried a violent tier1).
+
+**Do not re-derive `has_violence` from typology + intensity** — that rule (which an earlier
+revision of this file documented) disagrees with the code on every NEG row.
 
 **Do not remove it.** AI teams use it for binary baseline models and stratified train/val/test
-splits. Removing it forces every downstream user to re-derive it from `violence_typology` and
-`max_intensity` — which introduces inconsistency risk.
+splits. Removing it forces every downstream user to re-derive it — and the only correct rule
+is the events-based one above.
 
 ### The taxonomy is the ground truth
 
@@ -84,10 +94,15 @@ always appear alongside `violence_typology`, `tier1_category`, `tier2_subtype`,
 | Sample rate | 16 kHz |
 | Channels | Mono |
 | Bit depth | 16-bit PCM WAV |
-| Peak level | −1.0 dBFS (peak-normalized) |
+| Peak normalization target | `−2.0 dBFS` default (configurable, range `[−12.0, −1.5]`) |
+| Safety limiter ceiling | `≤ −1.0 dBFS` (no-op in normal flow) |
+| `preprocessing_applied.normalized_dbfs` | measured peak (lands near target ± preprocessor slop) |
+| `generation_metadata.loudness_target_peak_dbfs` | configured target |
 | Silence padding | ≥ 0.5 s before and after target speech |
 | Language | Hebrew (he-IL) |
 
+See [spec §3](https://github.com/DataHackIL/SynthBanshee/blob/main/docs/spec.md#3-audio-format-requirements) and the §5.1 field notes for the measured-vs-target split.
+
 ---
 
 ## How clips get here

diff --git a/DELIVERIES.md b/DELIVERIES.md
@@ -6,7 +6,8 @@ clip counts, prosody QA results, known limitations, and the SynthBanshee commit
 | # | Date | Slug | Project | Tier | Clips | Duration | Typologies | Pipeline milestone | Status | PR |
 |---|------|------|---------|------|------:|------:|------------|-------------------|--------|-----|
 | 001 | 2026-04-15 | [debug-run-1](deliveries/001-debug-run-1/notes.md) | she_proves | A | 1 | 2m 36s | IT | v1 baseline (pre-V3) | superseded | [#1](https://github.com/DataHackIL/avdp-synth-corpus/pull/1) |
-| 002 | 2026-04-15 | [m2a-wettest](deliveries/002-m2a-wettest/notes.md) | she_proves | A | 8 | ~17m | SV, IT, NEG, NEU | M2a SSML prosody | provisional | [#2](https://github.com/DataHackIL/avdp-synth-corpus/pull/2) |
+| 002 | 2026-04-15 | [m2a-wettest](deliveries/002-m2a-wettest/notes.md) | she_proves | A | 8 | ~17m | SV, IT, NEG, NEU | M2a SSML prosody | superseded | [#2](https://github.com/DataHackIL/avdp-synth-corpus/pull/2) |
+| 003 | 2026-05-12 | [multi-project-multi-voice](deliveries/003-multi-project-multi-voice/notes.md) | she_proves + elephant | A + B | 20 | ~42m | SV, IT, NEG, NEU | M2a + post-#78 / #102 / #103 / #105 / #106 | provisional | [#TBD](https://github.com/DataHackIL/avdp-synth-corpus/pulls) |
 
 ## Status definitions
 

diff --git a/README.md b/README.md
@@ -51,9 +51,10 @@ Every `.wav` must have a matching `.txt`, `.json`, and `.jsonl`. A clip without
 
 ## Clip ID and filename conventions
 
-- All filenames are **ASCII only**, lowercase, no spaces.
-- Format: `{scene_id}_{take_number:02d}` — e.g. `sp_it_a_0001_00`.
-- Speaker directory names match the `speaker_id` field in the clip's `.json` metadata.
+- All filenames (and filesystem path components) are **ASCII only**, lowercase, no spaces.
+- Format: `{scene_id_lower}_{take_number:02d}` — e.g. `sp_it_a_0001_00`. The same id appears uppercase in YAML `scene_id`.
+- The on-disk speaker directory is `speaker_id.lower()` of the scene's first listed speaker. The `speakers[].speaker_id` *value* in the `.json` stays uppercase (`AGG_M_30-45_001`); only the directory name is lowercase (`agg_m_30-45_001/`).
+- **Single source of truth for per-surface casing rules:** [SynthBanshee `docs/spec.md` §2.5 — Identifier casing (per surface)](https://github.com/DataHackIL/SynthBanshee/blob/main/docs/spec.md#25-filename-constraints).
 - **No Hebrew text in filenames or JSON keys/values** — Hebrew belongs in `.txt` transcript files only.
 
 ---
@@ -68,7 +69,13 @@ Labels follow a three-level hierarchy defined in `configs/taxonomy.yaml` in the
 | Tier 1 category (event-level) | `tier1_category` | `PHYS`, `VERB`, `DIST`, `ACOU`, `EMOT`, `NONE` |
 | Tier 2 subtype (event-level) | `tier2_subtype` | `VERB_THREAT`, `DIST_SCREAM`, `PHYS_HARD` |
 
-`has_violence` is a **derived convenience field** computed from the hierarchical taxonomy (`violence_typology`, `violence_categories`, `max_intensity`). It is provided for fast filtering and baseline modelling. The taxonomy columns are the ground truth — `has_violence` must never be the only label used for training.
+`has_violence` is a **derived convenience field** computed from the strong-label events, not from typology or intensity. The rule is pinned in [SynthBanshee `docs/spec.md` §5.1](https://github.com/DataHackIL/SynthBanshee/blob/main/docs/spec.md#51-per-clip-metadata-json) and lives in `synthbanshee/labels/generator.py`:
+
+```python
+has_violence = any(e.tier1_category != "NONE" for e in events)
+```
+
+This means `NEG` (Negative / Confusor) clips are correctly `has_violence: false` even at `max_intensity ≥ 3` — by definition NEG is "acoustically intense but non-violent" so every event lands `tier1_category: "NONE"`. Do **not** re-derive `has_violence` from typology + intensity alone; you will disagree with the data on every NEG row. The taxonomy columns are the ground truth — `has_violence` is for fast filtering and baseline modelling only, never the sole training label.
 
 Intensity is scored 1–5 per turn:
 
@@ -89,23 +96,16 @@ All clips must conform to:
 - **Sample rate:** 16 kHz
 - **Channels:** Mono
 - **Bit depth:** 16-bit PCM
-- **Peak level:** ≤ −1.0 dBFS (peak-limited; per-turn RMS targeting preserves within-scene loudness trajectory)
+- **Peak normalization:** target `−2.0 dBFS` (configurable, range `[−12.0, −1.5]`) via single global gain, then safety limiter at `≤ −1.0 dBFS`. The measured peak lands in `preprocessing_applied.normalized_dbfs`; the configured target lands in `generation_metadata.loudness_target_peak_dbfs`. See [spec §3](https://github.com/DataHackIL/SynthBanshee/blob/main/docs/spec.md#3-audio-format-requirements) and §5.1 field notes.
 - **Silence padding:** ≥ 0.5 s ambient baseline before and after target speech
 
 ---
 
 ## Pipeline versions and data quality
 
-Clips carry a `generation_metadata` block in their `.json` file (from pipeline v3 onward) recording which pipeline version, TTS backend, voice, and prosody controller produced them. Older clips (v1) lack this block.
-
-**Known limitations of v1 clips (pre-V3 pipeline):**
-- Near-zero amplitude variation across intensity levels
-- VIC voice baseline F0 too high (~215 Hz; rises to child-voice range at intensity 5)
-- Hebrew gender-agreement errors in aggressor→victim address forms
-- No inter-turn overlap or interruption
-- Missing creaky/breathy phonation in victim voice at high intensity
+Clips carry a `generation_metadata` block in their `.json` file when the generator recorded pipeline provenance (`pipeline_version`, `tts_backend` per speaker, `voice_family` per speaker, `mix_mode_used`, `normalization_strategy`, `loudness_target_peak_dbfs`, `breathiness_applied`, `effective_prosody_caps`). Older clips may have it as `null` — treat absence as "unknown", not as failure. See [spec §5.1](https://github.com/DataHackIL/SynthBanshee/blob/main/docs/spec.md#51-per-clip-metadata-json) field notes.
 
-v1 clips are suitable for **pipeline smoke-tests and label schema development only**. They should not be used as primary model training data for intensity 3–5 turns. The V3 audio pipeline redesign addressing these issues is documented in `docs/audio_generation_v3_design.md` in the SynthBanshee repo.
+**Per-delivery quality posture lives in `deliveries/{slug}/notes.md`.** Each delivery records the SynthBanshee commit, milestone state, prosody / acoustic QA findings, and any known limitations specific to that batch. Consumer teams reading the corpus should always start from the delivery notes for the clips they're working with rather than assuming a single global quality bar.
 
 ---
 

diff --git a/assets/scripts/30c75454a0582e778273f52206db321369fd41f78975c9deef61a3df1fdbf0f1.json b/assets/scripts/30c75454a0582e778273f52206db321369fd41f78975c9deef61a3df1fdbf0f1.json
@@ -0,0 +1,150 @@
+{
+  "turns": [
+    {
+      "emotional_state": "neutral",
+      "intensity": 1,
+      "pause_before_s": 0.0,
+      "phrase_hints": [],
+      "speaker_id": "BEN_M_40-55_003",
+      "text": "\u05e9\u05dc\u05d5\u05dd, \u05d0\u05e0\u05d9 \u05d1\u05df \u05d0\u05dc\u05d5\u05df. \u05d4\u05d2\u05e9\u05ea\u05d9 \u05e2\u05e8\u05e2\u05d5\u05e8 \u05e2\u05dc \u05d3\u05d7\u05d9\u05d9\u05ea \u05d4\u05d1\u05e7\u05e9\u05d4 \u05e9\u05dc\u05d9 \u05dc\u05d2\u05de\u05dc\u05ea \u05d4\u05d1\u05d8\u05d7\u05ea \u05d4\u05db\u05e0\u05e1\u05d4, \u05d5\u05d0\u05de\u05e8\u05d5 \u05dc\u05d9 \u05dc\u05d1\u05d5\u05d0 \u05d4\u05d9\u05d5\u05dd \u05dc\u05e7\u05d1\u05dc \u05ea\u05e9\u05d5\u05d1\u05d4."
+    },
+    {
+      "emotional_state": "calm",
+      "intensity": 1,
+      "pause_before_s": 0.5,
+      "phrase_hints": [],
+      "speaker_id": "SW_F_30-45_001",
+      "text": "\u05e9\u05dc\u05d5\u05dd, \u05d1\u05df. \u05e9\u05d1\u05d9 \u05d1\u05d1\u05e7\u05e9\u05d4. \u05ea\u05df \u05dc\u05d9 \u05e8\u05d2\u05e2 \u05dc\u05e4\u05ea\u05d5\u05d7 \u05d0\u05ea \u05d4\u05ea\u05d9\u05e7 \u05e9\u05dc\u05da \u05d1\u05de\u05d7\u05e9\u05d1. \u05db\u05df, \u05d0\u05e0\u05d9 \u05e8\u05d5\u05d0\u05d4 \u05e9\u05d4\u05d2\u05e9\u05ea \u05e2\u05e8\u05e2\u05d5\u05e8 \u05dc\u05e4\u05e0\u05d9 \u05db\u05e9\u05dc\u05d5\u05e9\u05d4 \u05e9\u05d1\u05d5\u05e2\u05d5\u05ea."
+    },
+    {
+      "emotional_state": "sad",
+      "intensity": 2,
+      "pause_before_s": 0.3,
+      "phrase_hints": [],
+      "speaker_id": "BEN_M_40-55_003",
+      "text": "\u05e0\u05db\u05d5\u05df, \u05e9\u05dc\u05d5\u05e9\u05d4 \u05e9\u05d1\u05d5\u05e2\u05d5\u05ea \u05e9\u05d0\u05e0\u05d9 \u05de\u05d7\u05db\u05d4. \u05d0\u05d9\u05df \u05dc\u05d9 \u05db\u05e1\u05e3 \u05d1\u05db\u05dc\u05dc, \u05d7\u05d9\u05db\u05d9\u05ea\u05d9 \u05d2\u05dd \u05e9\u05e2\u05d4 \u05d5\u05d7\u05e6\u05d9 \u05d1\u05d7\u05d5\u05e5 \u05d1\u05d0\u05d5\u05dc\u05dd \u05d4\u05d4\u05de\u05ea\u05e0\u05d4. \u05d0\u05d6 \u05de\u05d4 \u05d9\u05e6\u05d0 \u05de\u05d6\u05d4?"
+    },
+    {
+      "emotional_state": "calm",
+      "intensity": 2,
+      "pause_before_s": 0.8,
+      "phrase_hints": [],
+      "speaker_id": "SW_F_30-45_001",
+      "text": "\u05d0\u05e0\u05d9 \u05de\u05d1\u05d9\u05e0\u05d4 \u05e9\u05d6\u05d4 \u05de\u05ea\u05e1\u05db\u05dc, \u05d5\u05d0\u05e0\u05d9 \u05de\u05e6\u05d8\u05e2\u05e8\u05ea \u05e2\u05dc \u05d4\u05d4\u05de\u05ea\u05e0\u05d4. \u05d0\u05e0\u05d9 \u05e8\u05d5\u05d0\u05d4 \u05db\u05d0\u05df \u05d1\u05de\u05e1\u05de\u05db\u05d9\u05dd \u05e9\u05d4\u05d5\u05d5\u05e2\u05d3\u05d4 \u05d1\u05d7\u05e0\u05d4 \u05d0\u05ea \u05d4\u05e2\u05e8\u05e2\u05d5\u05e8 \u05e9\u05dc\u05da, \u05d0\u05d1\u05dc \u05dc\u05e6\u05e2\u05e8\u05d9 \u05d4\u05d1\u05e7\u05e9\u05d4 \u05e0\u05d3\u05d7\u05ea\u05d4 \u05e9\u05d5\u05d1. \u05d4\u05e1\u05d9\u05d1\u05d4 \u05d4\u05d9\u05d0 \u05e9\u05dc\u05e4\u05d9 \u05d4\u05e0\u05ea\u05d5\u05e0\u05d9\u05dd \u05d1\u05de\u05e2\u05e8\u05db\u05ea, \u05d9\u05e9 \u05d4\u05db\u05e0\u05e1\u05d4 \u05e8\u05e9\u05d5\u05de\u05d4 \u05e2\u05dc \u05e9\u05de\u05da \u05e9\u05e2\u05d5\u05d1\u05e8\u05ea \u05d0\u05ea \u05d4\u05e1\u05e3."
+    },
+    {
+      "emotional_state": "angry",
+      "intensity": 3,
+      "pause_before_s": 0.2,
+      "phrase_hints": [
+        {
+          "char_end_original": 15,
+          "char_start_original": 0,
+          "hint": "stress",
+          "phrase_id": "t4_p0"
+        },
+        {
+          "char_end_original": 107,
+          "char_start_original": 78,
+          "hint": "stress",
+          "phrase_id": "t4_p1"
+        }
+      ],
+      "speaker_id": "BEN_M_40-55_003",
+      "text": "\u05de\u05d4?! \u05e0\u05d3\u05d7\u05d4 \u05e9\u05d5\u05d1?! \u05d0\u05d9\u05d6\u05d5 \u05d4\u05db\u05e0\u05e1\u05d4? \u05d0\u05d9\u05df \u05dc\u05d9 \u05e9\u05d5\u05dd \u05d4\u05db\u05e0\u05e1\u05d4! \u05d0\u05e0\u05d9 \u05de\u05d5\u05d1\u05d8\u05dc \u05db\u05d1\u05e8 \u05e9\u05de\u05d5\u05e0\u05d4 \u05d7\u05d5\u05d3\u05e9\u05d9\u05dd! \u05d4\u05d1\u05d0\u05ea\u05d9 \u05d0\u05ea \u05db\u05dc \u05d4\u05e0\u05d9\u05d9\u05e8\u05d5\u05ea, \u05d0\u05ea \u05db\u05dc \u05d4\u05d0\u05d9\u05e9\u05d5\u05e8\u05d9\u05dd, \u05d5\u05d0\u05ea\u05dd \u05e9\u05d5\u05d1 \u05d3\u05d5\u05d7\u05d9\u05dd \u05d0\u05d5\u05ea\u05d9?!"
+    },
+    {
+      "emotional_state": "calm",
+      "intensity": 3,
+      "pause_before_s": 0.5,
+      "phrase_hints": [
+        {
+          "char_end_original": 40,
+          "char_start_original": 0,
+          "hint": "slow",
+          "phrase_id": "t5_p0"
+        }
+      ],
+      "speaker_id": "SW_F_30-45_001",
+      "text": "\u05d1\u05df, \u05d0\u05e0\u05d9 \u05e9\u05d5\u05de\u05e2\u05ea \u05d0\u05d5\u05ea\u05da \u05d5\u05d0\u05e0\u05d9 \u05e8\u05d5\u05d0\u05d4 \u05e9\u05d0\u05ea\u05d4 \u05d1\u05de\u05e6\u05d1 \u05e7\u05e9\u05d4. \u05d1\u05d5\u05d0 \u05e0\u05e0\u05e1\u05d4 \u05dc\u05d1\u05d3\u05d5\u05e7 \u05d0\u05ea \u05d6\u05d4 \u05d1\u05d9\u05d7\u05d3. \u05d9\u05d9\u05ea\u05db\u05df \u05e9\u05d9\u05e9 \u05d8\u05e2\u05d5\u05ea \u05d1\u05de\u05e2\u05e8\u05db\u05ea, \u05d5\u05d0\u05dd \u05db\u05df, \u05d0\u05e4\u05e9\u05e8 \u05dc\u05d4\u05d2\u05d9\u05e9 \u05d4\u05e9\u05d2\u05d4 \u05e0\u05d5\u05e1\u05e4\u05ea \u05e2\u05dd \u05de\u05e1\u05de\u05db\u05d9\u05dd \u05de\u05e2\u05d5\u05d3\u05db\u05e0\u05d9\u05dd."
+    },
+    {
+      "emotional_state": "angry",
+      "intensity": 4,
+      "pause_before_s": 0.2,
+      "phrase_hints": [
+        {
+          "char_end_original": 83,
+          "char_start_original": 62,
+          "hint": "stress",
+          "phrase_id": "t6_p0"
+        },
+        {
+          "char_end_original": 131,
+          "char_start_original": 84,
+          "hint": "slow",
+          "phrase_id": "t6_p1"
+        }
+      ],
+      "speaker_id": "BEN_M_40-55_003",
+      "text": "\u05d4\u05e9\u05d2\u05d4 \u05e0\u05d5\u05e1\u05e4\u05ea?! \u05e2\u05d5\u05d3 \u05e9\u05dc\u05d5\u05e9\u05d4 \u05e9\u05d1\u05d5\u05e2\u05d5\u05ea \u05e9\u05dc \u05d4\u05de\u05ea\u05e0\u05d4?! \u05d0\u05e0\u05d9 \u05dc\u05d0 \u05d9\u05db\u05d5\u05dc \u05dc\u05d7\u05db\u05d5\u05ea! \u05d4\u05d9\u05dc\u05d3\u05d9\u05dd \u05e9\u05dc\u05d9 \u05dc\u05d0 \u05d0\u05d5\u05db\u05dc\u05d9\u05dd! \u05d0\u05ea\u05dd \u05d9\u05d5\u05e9\u05d1\u05d9\u05dd \u05e4\u05d4 \u05de\u05d0\u05d7\u05d5\u05e8\u05d9 \u05d4\u05e9\u05d5\u05dc\u05d7\u05df, \u05dc\u05d0 \u05d0\u05db\u05e4\u05ea \u05dc\u05db\u05dd \u05de\u05d1\u05e0\u05d9 \u05d0\u05d3\u05dd! \u05db\u05dc \u05e4\u05e2\u05dd \u05e9\u05d0\u05e0\u05d9 \u05d1\u05d0 \u05dc\u05e4\u05d4 \u05d0\u05ea\u05dd \u05e9\u05d5\u05dc\u05d7\u05d9\u05dd \u05d0\u05d5\u05ea\u05d9 \u05d4\u05d1\u05d9\u05ea\u05d4 \u05d1\u05d9\u05d3\u05d9\u05d9\u05dd \u05e8\u05d9\u05e7\u05d5\u05ea!"
+    },
+    {
+      "emotional_state": "calm",
+      "intensity": 4,
+      "pause_before_s": 0.7,
+      "phrase_hints": [
+        {
+          "char_end_original": 27,
+          "char_start_original": 0,
+          "hint": "slow",
+          "phrase_id": "t7_p0"
+        }
+      ],
+      "speaker_id": "SW_F_30-45_001",
+      "text": "\u05d1\u05df, \u05d0\u05e0\u05d9 \u05de\u05d1\u05d9\u05e0\u05d4 \u05d0\u05ea \u05d4\u05db\u05e2\u05e1 \u05e9\u05dc\u05da, \u05d5\u05d0\u05e0\u05d9 \u05d1\u05d0\u05de\u05ea \u05e8\u05d5\u05e6\u05d4 \u05dc\u05e2\u05d6\u05d5\u05e8. \u05d1\u05d5\u05d0 \u05e0\u05d5\u05e8\u05d9\u05d3 \u05e7\u05e6\u05ea \u05d0\u05ea \u05d4\u05d8\u05d5\u05df, \u05db\u05db\u05d4 \u05e0\u05d5\u05db\u05dc \u05dc\u05de\u05e6\u05d5\u05d0 \u05e4\u05ea\u05e8\u05d5\u05df \u05d1\u05d9\u05d7\u05d3. \u05d0\u05e0\u05d9 \u05dc\u05e6\u05d9\u05d3\u05da."
+    },
+    {
+      "emotional_state": "threatening",
+      "intensity": 5,
+      "pause_before_s": 0.3,
+      "phrase_hints": [
+        {
+          "char_end_original": 117,
+          "char_start_original": 82,
+          "hint": "menace",
+          "phrase_id": "t8_p0"
+        },
+        {
+          "char_end_original": 167,
+          "char_start_original": 118,
+          "hint": "break_before",
+          "phrase_id": "t8_p1"
+        }
+      ],
+      "speaker_id": "BEN_M_40-55_003",
+      "text": "\u05dc\u05e6\u05d9\u05d3\u05d9?! \u05d0\u05ea \u05dc\u05e6\u05d9\u05d3\u05d9?! \u05d0\u05ea \u05d7\u05dc\u05e7 \u05de\u05d4\u05de\u05e2\u05e8\u05db\u05ea \u05d4\u05d6\u05d0\u05ea \u05e9\u05de\u05e8\u05de\u05e1\u05ea \u05d0\u05d5\u05ea\u05d9! \u05d0\u05dd \u05dc\u05d0 \u05ea\u05e1\u05d3\u05e8\u05d9 \u05dc\u05d9 \u05d0\u05ea \u05d4\u05d2\u05de\u05dc\u05d4 \u05d4\u05d9\u05d5\u05dd, \u05d0\u05e0\u05d9 \u05d0\u05d4\u05e4\u05d5\u05da \u05dc\u05da \u05d0\u05ea \u05db\u05dc \u05d4\u05de\u05e9\u05e8\u05d3 \u05d4\u05d6\u05d4. \u05d0\u05e0\u05d9 \u05d0\u05e9\u05d1\u05d5\u05e8 \u05d0\u05ea \u05d4\u05de\u05d7\u05e9\u05d1 \u05d4\u05d6\u05d4 \u05e2\u05dc \u05d4\u05e8\u05d0\u05e9 \u05e9\u05dc\u05da, \u05e9\u05d5\u05de\u05e2\u05ea?!"
+    },
+    {
+      "emotional_state": "fearful",
+      "intensity": 5,
+      "pause_before_s": 1.0,
+      "phrase_hints": [
+        {
+          "char_end_original": 24,
+          "char_start_original": 0,
+          "hint": "break_after",
+          "phrase_id": "t9_p0"
+        },
+        {
+          "char_end_original": 152,
+          "char_start_original": 94,
+          "hint": "slow",
+          "phrase_id": "t9_p1"
+        }
+      ],
+      "speaker_id": "SW_F_30-45_001",
+      "text": "\u05d1\u05df, \u05d0\u05e0\u05d9 \u05e6\u05e8\u05d9\u05db\u05d4 \u05e9\u05ea\u05e2\u05e6\u05d5\u05e8. \u05de\u05d4 \u05e9\u05d0\u05de\u05e8\u05ea \u05e2\u05db\u05e9\u05d9\u05d5 \u05d6\u05d5 \u05d0\u05d9\u05d5\u05de\u05d9\u05dd, \u05d5\u05d0\u05e0\u05d9 \u05dc\u05d0 \u05d9\u05db\u05d5\u05dc\u05d4 \u05dc\u05d4\u05de\u05e9\u05d9\u05da \u05d0\u05ea \u05d4\u05e9\u05d9\u05d7\u05d4 \u05db\u05db\u05d4. \u05d0\u05e0\u05d9 \u05e7\u05d5\u05e8\u05d0\u05ea \u05dc\u05d0\u05d7\u05e8\u05d0\u05d9 \u05d1\u05d9\u05d8\u05d7\u05d5\u05df. \u05d0\u05e0\u05d9 \u05de\u05d1\u05e7\u05e9\u05ea \u05de\u05de\u05da \u05dc\u05e9\u05d1\u05ea \u05d5\u05dc\u05d0 \u05dc\u05d4\u05ea\u05e7\u05e8\u05d1."
+    }
+  ]
+}